Data processing system

Abstract

The present application relates to a system for making legal modifications responsive to high efficiency, taking into account the reduction of illegal risk and the creation of new business opportunities. Using generative AI, relevant documents such as modified statutes and guidelines are automatically parsed, key points are extracted and abstracted, and the affected departments are automatically specified by checking with the business content database of each department within the company, and necessary application change plans, data correction plans, new business proposals, meeting materials, FAQs, and meeting scripts are automatically generated for each department. In addition, chatbot-based Q&A and progress, history, and notification management are also unified, characterized by comprehensive and rapid support for the entire process of legal modification response.

Description

Technical Field

[0001] The technology disclosed herein relates to a system. Background Technology

[0002] Patent Document 1 discloses a personalized chatbot control method executed by at least one processor, the method comprising: receiving user speech; adding the user speech to a prompt containing instructions related to a chatbot role; encoding the prompt; and inputting the encoded prompt into a language model to generate chatbot speech in response to the user speech.

[0003] Patent document 1: Japanese Patent Application Publication No. 2022-180282.

[0004] The problem this invention aims to solve is that every time a law is amended, each department within a company must grasp and analyze the content of the amended law, assess its impact on business, explore necessary response strategies, and individually and manually conduct a series of responses across the company, including notification, application changes, data revisions, implementation briefings, and Q&A sessions. This process is extremely time-consuming and energy-intensive, prone to omissions and legal risks, and fails to fully utilize new business opportunities arising from the legal amendment. Specifically, the problems include: the cumbersome work of collecting, summarizing, and analyzing diverse information such as amended law provisions, guidelines, public opinions, and legislative discussion records; the difficulty in accurately grasping the relevance of each department's business content to the legal amendment content to specifically impact the affected departments; the difficulty in quickly and accurately implementing necessary application changes and data revisions, preparing notification materials and FAQs, creating briefing scripts, and answering post-briefing questions for each affected department; furthermore, the discovery and proposal of new business opportunities arising from legal amendments are delayed, hindering proactive responses to legal amendments; and the difficulty in proposing specific response strategies tailored to the actual application and management situations within the affected departments, making it difficult to eliminate legal risks. To address all these challenges, the present invention aims to utilize generative AI to automate and streamline the entire process of legal amendment responses, achieving a balance between defensive and proactive responses. Summary of the Invention

[0005] The means for addressing the issues involved in this invention constitute a system supporting responses to legal amendments. This system includes a legal amendment information acquisition unit, an affected department determination unit, and a response plan generation unit. The legal amendment information acquisition unit has the function of automatically analyzing multiple legal amendment-related documents, such as amended legal provisions, guidelines, public opinions, and legislative discussion records, using generative AI, and extracting and summarizing the main points of the amendments, the amendment points, the implementation period, the new rules to be followed, and the impact on the industry. The affected department determination unit has the function of using generative AI to compare the business content database of multiple departments registered within the company with the key points of the legal amendments extracted by the legal amendment information acquisition unit, comparing the relevance of each department's business content to the legal amendment content by keywords or business processes, automatically determining the departments affected by the legal amendments, and prioritizing the departments with the most affected departments. The high-impact department's functions include: the response solution generation department, which uses generative AI to propose application change plans, data correction plans, and amendments to existing application manuals and internal company regulations for each department identified as affected by legal changes. It also proposes new business opportunities that may arise due to the legal changes. Furthermore, it automatically generates notification materials, FAQ drafts, briefing scripts, and time allocation plans for each department regarding the legal changes. A chatbot based on generative AI also provides real-time answers to questions raised after the briefing. This structure unifies and efficiently automates the entire process of responding to legal changes, reducing the risk of omissions and violations while promoting the discovery and proposal of new business opportunities arising from legal changes. It is a means of achieving a balance between defensive and proactive responses. Attached Figure Description

[0006] Figure 1 This is a conceptual diagram illustrating an example of the configuration of a data processing system according to the first embodiment.

[0007] Figure 2 This is a conceptual diagram illustrating an example of the main functions of the data processing apparatus and smart device according to the first embodiment.

[0008] Figure 3 This is a conceptual diagram illustrating an example of the data processing system configuration in the second embodiment.

[0009] Figure 4 This is a conceptual diagram illustrating an example of the main functions of the data processing device and smart glasses according to the second embodiment.

[0010] Figure 5 This is a conceptual diagram illustrating an example of the data processing system configuration in the third embodiment.

[0011] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing device and head-mounted terminal according to the third embodiment.

[0012] Figure 7 This is a conceptual diagram illustrating an example of the data processing system configuration in the fourth embodiment.

[0013] Figure 8 This is a conceptual diagram illustrating an example of the functions of the main parts of the data processing device and robot according to the fourth embodiment.

[0014] Figure 9 It represents an emotion graph that maps multiple emotions.

[0015] Figure 10 It represents an emotion graph that maps multiple emotions.

[0016] Explanation of reference numerals in the attached figures Data processing systems 10, 210, 310, and 410 12 Data processing device 14 Smart devices 214 Smart Glasses 314 Head-mounted terminal 414 Robot. Detailed Implementation

[0017] Hereinafter, an example of an implementation of the system involved in this disclosure will be described with reference to the accompanying drawings.

[0018] First, let's explain the terms used in the following description.

[0019] In the following embodiments, the processor (hereinafter referred to as "processor") can be a single computing device or a combination of multiple computing devices. Furthermore, a processor can be a single computing device or a combination of multiple computing devices. Examples of computing devices include CPU (Central Processing Unit), GPU (Graphics Processing Unit), GPGPU (General-Purpose computing on Graphics Processing Units), APU (Accelerated Processing Unit), etc.

[0020] In the following implementation, the labeled RAM (Random Access Memory) is a memory that temporarily stores information and is used by the processor as working memory.

[0021] In the following embodiments, the labeled memory is one or more non-volatile storage devices used to store various programs and various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), disk (e.g., hard disk) or magnetic tape, etc.

[0022] In the following implementation, the labeled Communication I / F (Interface) is an interface that includes a communication processor and an antenna, etc. The Communication I / F is responsible for communication between multiple computers. Examples of communication standards applicable to the Communication I / F include wireless communication standards such as 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), or Bluetooth (registered trademark).

[0023] In the following implementation, "A and / or B" is synonymous with "at least one of A and B". That is, "A and / or B" means that it can be only A, only B, or a combination of A and B. Furthermore, in this specification, when "and / or" connects more than three items, the same approach as "A and / or B" applies.

[0024] [First Implementation] Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0025] like Figure 1 As shown, the data processing system 10 includes a data processing device 12 and an intelligent device 14. An example of the data processing device 12 is a server.

[0026] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0027] The smart device 14 includes a computer 36, a receiver 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the receiver 38, output device 40, camera 42, and communication I / F 44 are also connected to the bus 52.

[0028] The receiving device 38 includes a touchscreen 38A and a microphone 38B, etc., for receiving user input. The touchscreen 38A receives user input generated by contact with an indicator (e.g., a pen or finger). The microphone 38B receives user input generated by sound by detecting the user's voice. The control unit 46A in the processor 46 sends data representing user input received through the touchscreen 38A and microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data representing the user input.

[0029] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting data in a form perceptible to the user 20 (e.g., sound and / or text). The display 40A displays visual information such as text and images according to instructions from the processor 46. The speaker 40B outputs sound according to instructions from the processor 46. The camera 42 is a small digital camera equipped with an optical system such as a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0030] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for sending and receiving various information between processor 46 and processor 28 via network 54.

[0031] Figure 2 An example of the main functions of the data processing device 12 and the smart device 14 is shown.

[0032] like Figure 2 As shown, in the data processing apparatus 12, specific processing is performed by the processor 28. A specific processing program 56 is stored in the memory 32. The specific processing program 56 is an example of a "program" as understood in this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 based on the specific processing program 56 executed on the RAM 30.

[0033] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0034] In the smart device 14, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The receive / output program 60 is used in conjunction with the data processing system 10 and the specific processing program 56. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart device 14 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0035] In addition, other devices besides the data processing device 12 may also have a data generation model 58.

[0036] For example, a server device (e.g., a generation server) may have a data generation model 58.

[0037] In this case, the data processing device 12 obtains the processing results (prediction results, etc.) using the data generation model 58 by communicating with a server device that has the data generation model 58. Furthermore, the data processing device 12 can be a server device or a user-owned terminal device (e.g., a mobile phone, robot, home appliance, etc.). Next, an example of the processing performed by the data processing system 10 of the first embodiment will be described.

[0038] (Example 1) The specific processing flow in Example 1 will be described below. Each part of the system described below is implemented by the data processing device 12 and the smart device 14. Furthermore, the data processing device 12 is referred to as the "server" and the smart device 14 is referred to as the "terminal".

[0039] (Implementation methods for carrying out the present invention) The embodiments of the present invention will be described in more detail below. The present invention is a system for supporting responses to legal amendments, comprising a server and a terminal. The server includes a legal amendment information acquisition unit, an affected department determination unit, a response plan generation unit, a database management unit, a generative AI execution unit, a log management unit, a notification management unit, and a history management unit. The terminal includes a user interface unit, a data viewing and editing unit, a chatbot response unit, a notification receiving unit, and an operation log sending unit. Specifically, this system employs an advanced information processing architecture that enables multiple computing nodes to work together in a distributed computing environment. The server is installed as a physical or virtual computer resource equipped with a multi-core processor and large-capacity volatile memory. The server contains a parallel computing cluster using a graphics processing unit. This parallel computing cluster accelerates the inference processing of large-scale language models. The terminal performs encrypted communication with the server via a network interface. The system converts the unstructured data of legal amendments into structured data. The system uses the converted structured data to quantitatively assess the impact on various departments within the enterprise. Based on the assessment results, the system automatically generates specific response steps. This system replaces the cognitive process of document reading and understanding the scope of impact performed by human legal professionals with natural language processing algorithms in a high-dimensional vector space. This results in a significant leap in processing speed. The system physically eliminates omissions in responding to legal changes caused by human error. This system improves the accuracy of compliance management. This system is particularly useful in industries subject to stringent legal regulations, such as financial institutions and healthcare facilities. The system containerizes each processing unit as a microservice architecture. The system dynamically scales the computing resources of each processing unit based on load. This prevents system response delays even under continuous large-scale legal changes.

[0040] First, the Legal Amendment Information Acquisition Department operates on a server. For example, legal counsel, general affairs officers, information management personnel, and foreign affairs officers can upload legal amendment-related documents from their terminals, including amended legal provisions, guidelines, public opinions, legislative discussion records, interpretations from the Executive Yuan, insights from industry groups, reference case law, administrative guidance documents, and various news reports. Uploads can be in any file format (PDF, Word, text, HTML, etc.). Furthermore, documents can be automatically retrieved from external legal databases or government websites via API linkage or web scraping. The uploaded or automatically retrieved document data is sent to the server and received by the Legal Amendment Information Acquisition Department. Specifically, this department executes a data preprocessing pipeline that converts unstructured data in various formats into a unified machine-readable format. It uses optical character recognition (OCR) algorithms to extract text data from image-based documents. The department performs lexical and syntactic analysis on the extracted text data. Finally, it removes unwanted symbols and whitespace that constitute noise. This Legal Amendment Information Acquisition Department periodically sends requests to external legal databases using the Hypertext Transfer Protocol (HTTP). It extracts discrepancies regarding legal amendments from the JSON or XML data received in response. The department uses web crawlers to parse the Document Object Model (DOM) tree of administrative agency websites. It automatically collects text elements surrounded by specific tag structures. It converts the collected text data into a sequence of tags in a one-dimensional array that can be input into a natural language processing (NLP) model. It divides the tag sequence into fixed-length blocks. It assigns positional encoding to each block. This allows it to transmit data to the subsequent generative AI execution department while preserving the document's contextual information. By automating data collection, the department eliminates the need for manual website inspection. It prevents information omissions. It minimizes the time lag between acquiring legal amendment information and registering it with the system. This department demonstrates significant technological effectiveness in the legal departments of global companies requiring real-time compliance responses.

[0041] The Legal Amendment Information Acquisition Department and the Generative AI Implementation Department collaborate to analyze uploaded or acquired documents using natural language processing technology. For example, the generative AI automatically extracts elements from amended legal provisions such as "newly added obligations," "deleted provisions," "effective date," "strengthening of penalty provisions," "applicable exceptions," "transitional measures," "changes in definitions," and "newly established reporting obligations," generating summaries. From guidelines and public opinions, it extracts information such as "administrative interpretation guidelines," "key opinions from industry," "points of contention during legislation," "responses from the Executive Yuan," "practical considerations," and "past violations," summarizing the background and practical considerations of the amendments. For example, in the case of amendments to the Personal Information Protection Act, it extracts key points such as "strengthened recording obligations when providing information to third parties," "added reporting obligations in case of leakage," "changed the definition of anonymized processed information," and "new regulations regarding cross-border transfers." From the Executive Yuan's interpretation notices and industry group insights, it extracts practical guidance such as "the Executive Yuan's guidelines for responding to situations like A" and "industry groups' recommendations for the application of C." Specific examples, such as "a judgment of E was made in a case like D in the past," are extracted from referenced case law and administrative guidance documents. These summaries and extraction results are stored in the database management department on the server and used for subsequent processing. Specifically, this generative AI execution unit is equipped with a large-scale language model based on the Transformer architecture. This generative AI execution unit uses a self-attention mechanism to calculate the dependencies between words within the input token sequence. This generative AI execution unit receives a one-dimensional array of token sequences output from the legal amendment information acquisition unit as input data. This input data is, for example, represented as a 512-dimensional word embedding vector group. This generative AI execution unit applies a multi-layer encoder network to the input data. This generative AI execution unit generates a high-dimensional feature tensor reflecting the context of the legal amendment document. This generative AI execution unit inputs the feature tensor into a decoder network. This generative AI execution unit outputs the probability distribution of the labels of the elements that become extraction objects and their corresponding text spans. This output data is, for example, an array of real values ​​from zero to one representing the probability that each token belongs to a specific extraction element. This generative AI execution unit identifies labeled sequences with probabilities exceeding a predetermined threshold as extraction results. For example, when receiving a revised document on the Personal Information Protection Law as input, this generative AI execution unit extracts the string "reporting obligation in case of leakage" with a high probability. This generative AI execution unit converts the extracted elements into key-value structured data. This generative AI execution unit sends the structured data to the database management unit. This generative AI execution unit performs masked language modeling tasks using a large amount of legal documents and case data during pre-training. This generative AI execution unit optimizes the model's weight parameters using a cross-entropy error function. This generative AI execution unit performs supervised learning using correct answer data prepared by legal experts during fine-tuning.This generative AI execution unit accurately parses the uniquely difficult expressions and complex syntax of law. It completes document reading and summarizing tasks that would take humans hours in mere seconds. This generative AI execution unit delivers significant technological benefits, including faster information processing and improved extraction accuracy. It is useful not only for improving the operational efficiency of legal departments but also for quality management, such as preventing misunderstandings of legal amendments.

[0042] The Department of Impact Assessment is also implemented on the server. The server contains a database registering each department's business content and processes, areas of responsibility, past legal amendment response history, organizational charts, business process diagrams, lists of managed assets, lists of external clients, overviews of relevant systems, data categories used, various regulations and manuals, educational materials, past internal company notices, and points raised in past internal audits. The Department of Impact Assessment uses generative AI to cross-reference the key points of legal amendments extracted by the Legal Amendment Information Acquisition Department with each department's business content database. For example, if the Information Management Department is responsible for "Management and Use of Personal Information," "Database Management," "Management of External Clients," and "Access Permission Management," it is automatically listed as an affected department because it is directly related to amendments to the Personal Information Protection Act. If the Legal Department is responsible for "Contract Management," "Specific Commercial Transactions," "Intellectual Property Management," and "Litigation Response," it is extracted as an affected department when the Specific Commercial Transactions Act or Copyright Act is amended. The generative AI compares the relevance of each department's business content to the legal amendment content using keywords or business processes, prioritizing departments with high impact. For example, if a department's business content includes keywords such as "data utilization," "external outsourcing," "communication services," "advertising," "customer information management," "employee information management," "contract signing," "electronic signature," and "electronic settlement," it is determined to be affected by relevant legal amendments. Furthermore, by analyzing organizational charts and business process diagrams, it automatically detects situations where the scope of impact expands, such as business spanning multiple departments or collaborative business with external clients. Specifically, this Department Impact Determination Unit executes an algorithm that compares the key points of the legal amendments with the attribute information of each department in a high-dimensional vector space. This Department Impact Determination Unit obtains text data representing the business processes and managed assets of each department from the Database Management Unit. This Department Impact Determination Unit inputs the obtained text data into the article embedding model. This Department Impact Determination Unit generates, for example, a fixed-length vector of 768 dimensions representing the business content of each department. Similarly, this Department Impact Determination Unit inputs the text data representing the key points of the legal amendments into the article embedding model. This Department Impact Determination Unit generates a fixed-length vector of 768 dimensions representing the key points of the legal amendments. This Department Impact Determination Unit calculates the cosine similarity between the department vector and the legal amendment vector. This influence department determination unit identifies departments with a cosine similarity exceeding a pre-set threshold as influential departments. It also uses graph neural networks to analyze the network structure of organizational charts and business process diagrams. This unit constructs graph data with departments as nodes and business linkages as edges. It performs graph convolution operations. Even for departments with low direct correlation, this unit quantifies the ripple effect on departments closely linked in business processes. The unit generates an array of scores from zero to one hundred, representing the influence of each department, as output data.This Department of Impact Assessment outputs, for example, a score of 95 for the Information Management Department and 70 for the Sales Department. Based on these scores, the Department of Impact Assessment sorts the list of affected departments in descending order. This sorted list is then sent to the subsequent response generation department. This Department of Impact Assessment replaces the specific operations relying on human intuition and experience regarding the scope of impact with an objective evaluation process based on mathematical vector operations and graph analysis. This Department of Impact Assessment thus completely prevents the omission of specific affected departments. In large corporations with complex organizational structures, this Department of Impact Assessment demonstrates exceptional technical effectiveness in accurately and comprehensively assessing the scope of impact of specific legal changes.

[0043] The response solution generation department is also implemented on the server. For each department identified as affected, generative AI automatically generates specific response solutions based on legal amendments. For example, for the Information Management Department, it proposes changes such as "reassessing the process for using third-party records of personal information," "establishing new reporting procedures in case of leaks," "revising contract templates with external clients," and "revising access control regulations," as well as amendments to existing user manuals and internal company regulations. For the Legal Department, it proposes "adding new contract clauses," "revising the description of specific business transactions," "introducing new consent acquisition procedures," and "reassessing litigation risk assessments." For the General Affairs Department, it proposes "revising the shareholder meeting operation manual," "updating employee training materials," and "creating internal company notification documents." For the External Affairs Department, it proposes "revising the communication service usage agreement," "revising explanatory materials for external negotiations," and "changing the reporting procedures to administrative agencies." Specifically, this response solution generation department drives a deep learning model to perform conditional text generation tasks. This response solution generation department receives the identifiers of the affected departments and their impact scores as input from the affected department's assessment department. The Response Generation Department obtains existing application manuals and internal company regulations from the Database Management Department. It constructs a prompt string combining the key points of the legal amendment with the existing regulations. This prompt string is then input into the Generative AI Execution Department. The department parses the context of the input prompt string, identifies which part of the existing regulations conflicts with the legal amendment, and generates, autoregressively, specific corrective text to eliminate the conflicting parts. The department generates structured JSON data as output, containing information on the differences between the corrected and revised text. This output data includes elements such as the filename of the object to be corrected, the line number of the corresponding paragraph, strings to be deleted, and strings to be appended. Based on the generated structured data, the department automatically creates update patches for existing documents. The department also uses a vector search engine to retrieve past experiences with similar legal amendment responses. It obtains text data of past successful and unsuccessful cases and appends these past cases to the prompt string. This solution generation department thereby improves the feasibility and appropriateness of the generated solutions. It replaces the labor-intensive task of manually identifying and revising regulations with a computer-automated process. This solution generation department significantly reduces solution creation time. It provides powerful technical means to complete effective compliance responses within the limited timeframe leading up to the effective date of the legal amendment.

[0044] Furthermore, new business opportunities arising from legal amendments are also proposed by generative AI. For example, it suggests proactive new business solutions such as "launching new data utilization services," "developing consulting services to comply with the amended law," "developing services targeting new customer segments," "entering new markets through business partnerships with other companies," "introducing new advertising methods," and "providing automated services utilizing AI." In addition, it automatically generates informational materials regarding the legal amendments, draft FAQs, briefing scripts, and time allocation plans for each department. For example, it generates informational materials on "changes in shareholder meeting operations due to the amendments to the Company Law," as well as frequently asked questions and their answers for the General Affairs Department. The briefing script includes a suggested time allocation such as "10 minutes for a summary of the amendments, 15 minutes for explaining specific response strategies, and 5 minutes for Q&A." The draft FAQ automatically generates hypothetical questions and answers regarding "the scope of application of the new regulations," "the impact on existing business," "the specific steps for implementing the changes," "the necessity of internal company training," and "responses to external clients." Specifically, the response solution generation department extracts elements of relaxed regulations and newly established systems from the key points of the legal amendments. This department multiplies the extracted elements with data on the company's technological assets and customer base. The department executes reinforcement learning algorithms to explore potential business opportunities. It inputs specific prompts into the generative AI execution department to generate new business ideas. The generative AI execution department generates a multi-dimensional evaluation matrix as output data, containing the business concept, target layer, required resources, and expected revenue scale. The department also uses natural language generation technology to automatically construct presentation slides. It summarizes the text data of the legal amendments and response solutions in bulleted format. It assigns the summarized text to each slide. It generates an XML file containing slide layout information. In generating the presentation script, the department outputs text data with metadata including pronunciation marks and pause positions for easy processing by a text-to-speech system. In generating FAQs, the Solution Generation Department clusters frequently occurring problem patterns from a database of past internal consulting experience. For each cluster, the department generates responses based on legal amendments, focusing on representative questions. The output data consists of a two-dimensional array of question and response pairs. Through these automated processes, the Solution Generation Department dramatically reduces the computer data processing time spent on document preparation by the legal and planning departments. The Solution Generation Department goes beyond simply addressing risk in legal amendments; it achieves strategic information processing directly related to strengthening the company's competitiveness.This solution generation department is extremely useful as a decision support system for business planning and new business development departments.

[0045] Generative AI is essential in all major processes across departments. For example, in the Legal Amendment Information Acquisition Department, generative AI quickly parses a large number of legal amendment-related documents, extracting and summarizing key points. In the Impact Department Assessment Department, generative AI analyzes the relevance of each department's business content to the legal amendment content from multiple perspectives, scoring the impact and assigning priorities. In the Response Generation Department, generative AI automatically generates application change plans, data correction plans, and new business proposals tailored to each department's specific circumstances, while referencing past response examples and industry best practices. For instance, in cases of similar legal amendments in the past, it can refer to past response experiences to suggest the best response plan for the current amendment. Furthermore, generative AI learns from the past problem-solving experiences of those responsible in each department, as well as the Q&A content from past briefings, enabling it to provide more practical proposals and answers. Specifically, this system positions the generative AI execution department as the core inference engine of the entire system. This system manages inference requests from each processing department to the generative AI execution department via asynchronous message queues. This generative AI execution unit possesses an adapter module that dynamically switches a portion of the model's weight parameters based on various input tasks. This generative AI execution unit maintains optimized prompt templates for summarization, classification, and text generation tasks in memory. This generative AI execution unit executes an online learning pipeline for continuously learning from past responses and answers. This online learning pipeline collects feedback data from users. This online learning pipeline uses the feedback data to train a reward model. This online learning pipeline uses reinforcement learning to optimize the generative AI's output strategy. This generative AI execution unit thus achieves a self-evolving architecture where the longer the system is used, the higher the accuracy and appropriateness of the output. This generative AI execution unit receives text pairs of past answers as input data. This generative AI execution unit generates updated model weight tensors as output data. This system distributes the generative AI's inference processing across the parallel computing cores of the graphics processing unit. This system thus maximizes throughput. This system can perform flexible information processing with an understanding of context, even for unknown legal modification patterns that existing rule-based systems cannot handle. This system does not treat artificial intelligence technology as a mere auxiliary tool, but rather enables it to function as an autonomous computing entity controlling the data flow of the entire system. This allows the system to completely digitize complex enterprise-level compliance processes.

[0046] A database management department exists on the server, providing unified management of all data, including legal amendment-related documents, departmental business content, past response history, generated summaries, response plans, materials, FAQs, scripts, and Q&A. A log management department records the system's entire processing history, user operation history, and AI output history, which is used for future auditing and ensuring traceability. A notification management department automatically distributes notifications to all terminals regarding legal amendment information and response plans, material updates, briefing guides, and FAQ updates. A history management department manages past legal amendment response history and departmental response status in chronological order, serving as reference information for future legal amendments. Specifically, this database management department employs a hybrid data storage architecture combining relational and NoSQL databases. Structured business data and user information are stored in a relational database. Unstructured document files and generative AI output text are stored in a document-oriented NoSQL database. Each piece of data is calculated and assigned a unique hash value. A mechanism for detecting data tampering is provided. This log management department records all events occurring within the system along with timestamps in a time-series database. It also saves detailed log data, including input prompts to generative AI, output results, processing time spent on inference, and model version information used. This log management department enables the AI's judgments to be verified afterward. This notification management department uses a publish-subscribe model-based messaging protocol. It sends push notifications in real-time via WebSocket connections established with various terminals. It tracks the read status of notifications. It automatically escalates unread notifications. This history management department models the response status of each department as a state machine. It strictly manages the transitions between states: not initiated, in progress, awaiting approval, and completed. Through the collaboration of these management departments, the system ensures data integrity and availability. The system provides the ability to immediately submit accurate records for information disclosure requests from auditors or regulatory authorities. This system brings the crucial technological benefits of enhanced corporate governance and automated compliance documentation.

[0047] On the terminal side, a user interface unit is installed, allowing department heads to access the server and view and download generated summaries, lists of affected departments, response plans, informational materials, FAQs, and briefing scripts. For example, legal personnel can confirm on the terminal "What matters should our department address in this revision," "Which materials should be revised," and "What explanations should be given at the briefing." In the data viewing and editing department, generated materials can be used as is or supplemented and revised by department. For example, information management personnel can add descriptions of their department's unique usage to the AI-generated application manual revision plan. Specifically, this user interface unit is installed as a single-page application running on a web browser. It uses asynchronous communication to retrieve JSON-formatted data from the server. The retrieved data is bound to a virtual document object model. The user interface unit achieves high-speed rendering while minimizing screen redrawing. The generated summaries and response plans are rendered on the screen as a hierarchical tree structure. The user interface unit dynamically expands detailed information when the user clicks on a specific item. This document viewing and editing unit includes a rich text editor component. It provides a difference display function that shows the AI-generated corrections side-by-side with the original text. It captures keystrokes when the user makes supplementary corrections. It synchronizes changes to the server in real time. It implements an exclusivity control algorithm to prevent conflicts when multiple users are editing the same document simultaneously. To reduce the user's workload, this terminal can also integrate multimodal interfaces such as eye tracking and voice input. By providing an intuitive and responsive user interface, this system lowers the barrier to entry for non-technical legal and administrative personnel. This system significantly improves the efficiency of human-computer interaction. This system seamlessly integrates the final confirmation and correction process of AI-generated results by humans. This system maximizes business efficiency while avoiding the risks of full automation.

[0048] The chatbot response department is also implemented on the terminal. After briefings and in daily operations, when a user inputs a question on the terminal, the content is sent to the server. The generative AI on the server, referencing the question content, key legal amendments, departmental business content, and past Q&A history, generates the best answer and sends it back to the terminal. For example, for specific questions such as "Under what circumstances do the new record-keeping obligations apply?", "Where should the existing user manual be revised?", "At what time should notifications be sent to external clients?", and "Which part of the educational materials should be updated?", the AI ​​provides an immediate answer. Specifically, this chatbot response department provides a natural language-based conversational interface. The chatbot response department sends the user-input text string as a POST request using the Hypertext Transfer Protocol (HTTP) to the server. The generative AI execution department on the server employs a retrieval-enhanced generative architecture. This generative AI execution department vectorizes the input question string. This generative AI execution department performs a similarity search on the vector database. This generative AI execution department extracts text blocks related to the question, including key legal amendments, departmental business content, and past Q&A history. This generative AI execution unit combines extracted text blocks as contextual information into the question string. The combined prompt is then input into a large-scale language model. This generative AI execution unit generates accurate and non-illusionary response text based on the contextual information. This generative AI execution unit generates a JSON object as output data, containing the response text and links to reference documents used as the basis for the response. This chatbot response unit parses the received JSON object. This chatbot response unit displays the response text as a speech bubble-style user interface element. This chatbot response unit also adds hyperlinks to the reference documents. This system, through this conversational processing, eliminates the time users spend searching and reading vast amounts of manuals and legal amendments. This system accurately provides the necessary information. This system fully automates the company's internal help desk operations. This system delivers the significant technological benefit of dramatically reducing the human resource costs associated with consultation responses.

[0049] The notification receiving department receives notifications from the server in real time and notifies the responsible parties via pop-up windows, emails, chat messages, etc. For example, it can instantly receive notifications such as "New legal amendment information added," "This department has been selected as an affected department," "New response plan generated," "Guidelines for holding briefings," and "FAQ updated." The operation log sending department sends the operation history from the terminal to the server, uniformly managing the system's utilization and response progress. Specifically, this notification receiving department utilizes a persistent background process service thread. This notification receiving department works in conjunction with the push notification API provided by the terminal's operating system. This notification receiving department captures notification events from the server even when the web browser is closed. This notification receiving department parses the payload of the received notification data. This notification receiving department dynamically changes the display format of the notification based on flags indicating the importance of the notification. For example, in the case of a notification regarding a high-importance legal amendment response deadline, this notification receiving department displays a modal dialog box in the center of the screen, forcing the user to confirm. This operation log sending department collects detailed interaction data such as user mouse clicks, scrolling, keyboard input, and screen dwell time. This operation log sending unit stores collected data in local storage until a certain buffer size is reached. This operation log sending unit compresses data to reduce network load. This operation log sending unit asynchronously sends the compressed log data to the server in batches. Through these functions, this system ensures reliable information delivery to users. This system quantitatively visualizes users' system utilization. This system identifies bottlenecks in specific user interfaces through operation log analysis. This system constructs a data-driven feedback loop for continuous improvement of system availability. This system provides the foundational data for dashboard functionality that monitors the progress of compliance responses in real time.

[0050] Furthermore, even when multiple legal amendments occur simultaneously, the generative AI can independently analyze each amendment and automatically adjust the consistency of response plans for affected departments. For example, if the Personal Information Protection Act and the Telecommunications Business Act are amended simultaneously, it can suggest combined application change plans and data revision plans for departments affected by both amendments. Additionally, even when departments have different business processes and management systems, the generative AI learns each department's business processes and past application history to generate optimal response plans individually. For example, even with the same legal amendment, it can individually propose revisions to database management regulations to the Information Management Department, updates to employee training materials to the General Affairs Department, and revisions to external negotiation materials to the External Affairs Department. Specifically, this system employs a multi-threaded parallel processing architecture to handle multiple legal amendment events. The generative AI execution unit encodes the multiple input legal amendment documents as independent tensors. The generative AI execution unit generates feature vectors for each legal amendment. The generative AI execution unit combines multiple feature vectors and inputs them into a cross-attention mechanism. This generative AI execution unit detects the interactions and contradictions between legal amendments. For example, it receives vectors from both the Personal Information Protection Act and the Telecommunications Business Act as input. It extracts the common data protection requirements of both regulations. The unit outputs a single, optimized response vector that integrates repetitive response tasks. The response generation unit decodes this optimized vector. It generates textual data for composite application change plans. The unit takes departmental context as conditional variables. Using a conditional generative model, it generates department-specific outputs. For example, given context variables from the Information Management Department, it increases the probability of terms related to database management. Similarly, given context variables from the General Affairs Department, it increases the probability of terms related to education and training. Through this high level of contextual understanding and reasoning ability, the system minimizes the burden on enterprises dealing with complex and intertwined regulations. The system automatically generates actionable response plans tailored to the specific circumstances of each department. This system achieves a high level of technical effectiveness by simultaneously ensuring consistency and efficiency in compliance responses across the entire company.

[0051] Furthermore, for example, the progress management of legal amendment responses can also be unified by the system. When the responsible persons of each department input the response status from the terminal, the server's history management department calculates the progress, automatically extracts the departments that have not responded and the unfinished tasks, and sends reminder notifications from the notification management department. For example, reminders such as "User manual revision not completed" and "Explanation meeting materials confirmation not implemented" can be automatically distributed. Specifically, this history management department models the progress status as a directed acyclic graph, treating each response task as a node and the dependencies between tasks as edges. This history management department receives API requests for status updates from the terminal. This history management department updates the status flags of the corresponding tasks in the database. This history management department executes a graph search algorithm. This history management department identifies subsequent tasks that become executable due to the completion of preceding tasks. This history management department periodically performs batch processing. This history management department compares the current date and time with the deadline date and time set in each task. This history management department extracts unfinished tasks whose remaining time before the deadline is less than a predetermined threshold. This history management department sends the list of extracted tasks to the notification management department. This notification management department resolves the target address based on the identifier of the task's responsible person. This notification management department generates and sends reminder notification message payloads. This system automates schedule management, eliminating the management costs of project managers manually confirming status with various departments. This system detects task delays early. This system mathematically and systematically reduces the risk of compliance violations related to the absolute deadline of legal amendments. This system significantly improves the reliability and transparency of project management in large-scale organizations.

[0052] Furthermore, for example, some system functions can be opened to external clients and group companies to enhance the coordination of responses to legal amendments. External clients can input their company's response status from the terminal or refer to AI-generated data. For example, it can automatically generate "new contract templates for external clients" and "application change plans for group companies," and share them among relevant personnel. Specifically, this system adopts a multi-tenant architecture. This system provides data areas and access permissions logically separated by external clients and group companies. This system is equipped with a role-based access control mechanism. This system only allows external users to access tasks and data related to this company. The response plan generation department receives attribute information and contract content data from external clients as input. This response plan generation department uses generative AI to generate customized contract templates and application change plans specifically for external clients. This generative AI execution department generates text data with different security requirements and reporting obligations, for example, in cases where the scope of the entrusted business is system development and call center business. This system distributes the generated data to the terminals of external users via a secure file-sharing protocol. This system receives progress reports from external users via a standardized API. This system integrates it into the company's progress management chart. It manages legal modification responses across the entire supply chain on a single platform. It prevents delays and disagreements in communication across enterprises. It delivers a powerful technological effect by homogenizing compliance levels throughout the ecosystem and mitigating legal risks arising from external clients.

[0053] In this embodiment of the invention, generative AI is effectively utilized in the main processing unit on the server side, while user experience on the terminal side includes user interface and chatbot responses, notification reception, data editing, and progress management. This unifies and efficiently automates the entire process of responding to legal amendments, reducing the risk of omissions and legal violations while promoting the discovery and proposal of new business opportunities arising from legal amendments, achieving a balance between defensive and offensive responses. Specifically, this system functions as an information processing device that highly integrates natural language processing, high-dimensional vector operations, graph analysis, and distributed computing technologies. This system transforms existing legal compliance processes, which rely on the limits of human cognitive abilities, into scalable and reproducible algorithmic processing based on computer science. The system constructs a pipeline of processes, including structuring unstructured data, defining the scope of influence based on semantic similarity, and generating text that understands context. This system thereby minimizes data processing latency. Furthermore, this system does not limit the reasoning capabilities of generative AI to simple text generation but applies it as an advanced analytics engine for deriving corporate business strategies. This system takes changes in the external environment, such as legal amendments, as input data, and simultaneously generates both defensive tasks for risk mitigation and offensive tasks for maximizing profits as optimized output data. Through efficient utilization of hardware resources and optimized software algorithms, this system establishes a new paradigm for enterprise systems. Through fundamental improvements in information processing technology, this system creates crucial technological and economic benefits that ensure both sustainable growth and legal security for enterprises.

[0054] (System Composition) The system described in this embodiment includes a legal amendment information acquisition unit, an affected department determination unit, a response plan generation unit, a database management unit, a generative AI execution unit, a log management unit, a notification management unit, a history management unit, a user interface unit, a data viewing and editing unit, a chatbot response unit, a notification receiving unit, and an operation log sending unit. Specifically, each component of this system is installed as a software module running on physical or virtual hardware resources equipped with a central processing unit, random access memory, a solid-state drive, and a network interface card. The legal amendment information acquisition unit, affected department determination unit, response plan generation unit, database management unit, generative AI execution unit, log management unit, notification management unit, and history management unit, which are server-side components of this system, communicate with each other via a high-speed internal bus or a broadband local area network. The generative AI execution unit communicates directly with dedicated accelerator hardware specialized for tensor operations, processing large-scale matrix multiplication and addition operations at high speed. The user interface unit, data viewing and editing unit, chatbot response unit, notification receiving unit, and operation log sending unit, which are terminal-side components of this system, run on the processor of a client device such as a personal computer or smartphone. The terminal-side component interacts with the server-side component via a communication channel encrypted with Secure Sockets Layer (SSL). Through the coordinated action of these hardware and software components, the system executes computationally intensive natural language processing tasks in real time. The system employs a redundant configuration to ensure overall system availability and fault tolerance. It provides a robust architecture capable of continuing operation even in the event of a specific hardware node failure.

[0055] The Legal Amendment Information Acquisition Department is capable of receiving various legal amendment-related documents uploaded from terminals by legal, general affairs, information management, and foreign affairs personnel. These documents include amended legal provisions, guidelines, public opinions, legislative discussion records, interpretations from the Executive Yuan, insights from industry groups, reference cases, administrative guidance documents, and various news reports. It also automatically retrieves these documents from external legal databases or government websites via API linkage or web scraping. For example, it can receive amended legal provisions in PDF format, automatically retrieve guidelines from the Executive Yuan published in HTML format, and allow users to upload insights from industry groups distributed as Word files. Additionally, it can automatically collect and summarize opinions from public opinion solicitation pages. The Legal Amendment Information Acquisition Department then transmits the received or acquired documents to the Generative AI Execution Department, which generates prompts for parsing. For example, it generates prompts such as "Summarize the following legal amendments and extract key amendment points, implementation dates, added obligations, deleted provisions, strengthening of penalties, and transitional measures," "Summarize the Executive Yuan's interpretation policy, practical considerations, and impact on the industry from the following full-text guidelines," and "Summarize the main opinions and the Executive Yuan's responses from the following public opinion results," and inputs them into the generative AI. Specifically, this Legal Amendment Information Acquisition Department has a dedicated group of parsers for parsing various file formats. For portable document format files, it decodes the internal text stream and extracts text data while preserving layout information. For Hypertext Markup Language files, it constructs a document object model and extracts only the text corresponding to nodes in the main text. It normalizes the extracted text data by character encoding and removes unnecessary control characters. It stores the preprocessed text data in a buffer on memory. It counts the number of characters in the text data. When the number of characters exceeds the input token limit of the generative AI, the Information Acquisition Department of this Law Amendment executes a sliding window algorithm to segment the text into multiple blocks at semantic paragraph boundaries. The Information Acquisition Department dynamically combines the prompt text into each block. The Information Acquisition Department then formats the combined string into input data for the Generative AI Execution Department. This input data is, for example, a JSON-formatted payload containing system prompts representing the instructions and user prompts representing the text of the parsed object. The Information Acquisition Department sends this payload to the application programming interface of the Generative AI Execution Department. The Information Acquisition Department of this Law Amendment fully automates the entire process from data collection to preprocessing and prompt generation. The Information Acquisition Department of this Law Amendment eliminates manual data shaping work performed by humans. The Information Acquisition Department of this Law Amendment plays a crucial technical role in consistently supplying the optimal input data to maximize the accuracy and efficiency of subsequent AI processing.

[0056] The Department of Impact Assessment compares the business content, business processes, areas of responsibility, past legal amendment response history, organizational charts, business process diagrams, asset lists of managed entities, lists of external clients, overviews of relevant systems, data categories used, various regulations and manuals, educational materials, past internal company notices, and past internal audit findings of each department registered on the server with the key points of legal amendments extracted by the Department of Legal Amendment Information Acquisition, using generative AI analysis to determine the correlation. For example, if the Information Management Department, responsible for "Management and Use of Personal Information," "Database Management," "Management of External Clients," and "Access Permission Management," is directly related to amendments to the Personal Information Protection Law, it is automatically listed as an affected department. Similarly, the Legal Department, responsible for "Contract Management," "Specific Commercial Transactions," "Intellectual Property Management," and "Litigation Response," is identified as an affected department when the Specific Commercial Transactions Law or Copyright Law is amended. Furthermore, the General Affairs Department, responsible for "Shareholders' Meeting Operations," "Employee Education," and "Preparation of Internal Company Notices," becomes an affected department when the Company Law is amended. For example, departments responsible for "revising the regulations for the use of telecommunications services," "revising external negotiation materials," and "changing the procedures for submitting applications to administrative agencies" become affected departments when the Telecommunications Business Act is amended. The Department of Departmental Impact Assessment analyzes organizational charts and business process diagrams, automatically detecting even when the scope of impact expands to include business spanning multiple departments or collaborative business with external clients. For generative AI, prompts such as "Check the following key points of the legal amendment against the business content databases of each department to list the affected departments. Also extract highly relevant business processes and managed assets." are used. Specifically, this Department of Departmental Impact Assessment uses an advanced matching algorithm for multi-dimensional attribute data. The Department of Departmental Impact Assessment expands the attribute information of each department obtained from the database into a structured JSON object consisting of key-value pairs in storage. The Department of Departmental Impact Assessment combines the text data of the key points of the legal amendment with the JSON objects of each department to construct a composite prompt for input into the generative AI execution department. The generative AI execution department parses the input composite prompt. The generative AI execution department infers the semantic relationships between the legal elements contained in the key points of the legal amendment and the business processes and managed assets of each department. This generative AI execution unit generates structured data as output data, including identifiers of affected departments, confidence scores ranging from zero to one indicating the degree of correlation, and a list of specific business process names determined to be correlated. This affected department determination unit further constructs the business process diagram data as a directed graph in memory. This affected department determination unit uses departments determined by the generative AI to have a direct impact as starting nodes. This affected department determination unit uses a graph search algorithm to calculate the propagation of the impact to adjacent nodes.This Impact Department Determination Department automatically identifies the sales department as a secondary impact department, for example, when a system change in the information management department affects the sales department's customer response process. Through these processes, this Impact Department Determination Department logically and comprehensively identifies cross-departmental impacts that specific human stakeholders might easily overlook. This Impact Department Determination Department accurately visualizes compliance risks across the entire organization. This Impact Department Determination Department significantly improves the accuracy of scope definition in the early stages of legal amendment response projects.

[0057] The Response Development Department uses generative AI to automatically generate specific response plans based on legal amendments for each department identified as affected. For example, it proposes changes such as "reassessing the process for using third-party records of personal information," "establishing new reporting procedures in case of leaks," "revising contract templates with external clients," and "revising access control regulations," as well as amendments to existing user manuals and internal company regulations. For example, it proposes to the Legal Department "adding new contract terms," ​​"revising the description of specific business transactions," "introducing new consent acquisition procedures," and "reassessing litigation risk assessments." For example, it proposes to the General Affairs Department "revising the Shareholders' Meeting Operations Manual," "updating employee training materials," "creating internal company notification documents," and "redesigning internal company training programs." For example, it proposes to the External Affairs Department "revising the communication service usage agreement," "revising explanatory materials for external negotiations," "changing the reporting procedures to administrative agencies," and "conducting legal suitability assessments for new services." Furthermore, the generative AI also makes suggestions regarding new business opportunities that may arise due to legal amendments. For example, prompts could include "launch new data utilization services," "expand consulting services to comply with the amended law," "develop services for new customer segments," and "enter new markets through business collaborations with other companies." The system automatically generates informational materials, draft FAQs, briefing scripts, and time allocation plans for each department regarding legal amendments. Examples include generating informational materials on "changes in shareholder meeting operations due to amendments to the Company Law," frequently asked questions and their answers, and scripts such as "a 10-minute overview of the amendments, a 15-minute explanation of specific response strategies, and a 5-minute Q&A session." As examples of prompts for generative AI, examples could include: "Based on the following key points of the legal amendments and the department's business content, create necessary application change plans, data revision plans, revision plans for existing manuals, and new business proposals." or "Create a briefing script and draft FAQ based on the following content." Specifically, this response generation department applies prompting engineering techniques to maximize the reasoning capabilities of generative AI. The input prompts include format specifications, constraints, and instructions for the thought process in a phased manner. Following the input prompts, this generative AI execution department first performs a gap analysis between the key points of legal amendments and the department's business content. Next, it lists specific action items to bridge this gap. Finally, it structures these action items as application change plans and data revision plans. As output data, it generates a JSON array containing task names, priorities, roles, and specific revision text. In generating new business proposals, this response generation department crosses keywords related to regulatory easing resulting from legal amendments with keywords representing the company's core competencies.To encourage generative AI to generate creative ideas, the Solution Generation Department temporarily sets the model's temperature parameters to a higher level. In generating scripts for presentations, the Solution Generation Department automatically calculates the required time based on the number of characters in the text and the typical speaking speed. The Solution Generation Department dynamically adjusts the text components to converge within a specified time allocation. Through these advanced generation processes, the Solution Generation Department instantly outputs concrete response plans and deliverables that departments can immediately implement. The Solution Generation Department dramatically reduces the lead time for legal amendment responses from months to days. The Solution Generation Department provides powerful technological effects that improve corporate agility and fundamentally transform the productivity of the legal department.

[0058] The Database Management Department centrally manages all data related to legal amendments, departmental business content, past response records, generated summaries, response plans, materials, FAQs, scripts, Q&As, and other data. For example, it accumulates revised application manuals, briefing materials, FAQs, and Q&A records created during past legal amendments for future reference. It also records response records for multiple legal amendments occurring simultaneously, and the progress of each department's response. Furthermore, it manages collaboration records with external clients and group companies, facilitating information sharing among relevant personnel. It also records various notification and reminder records and utilizes them to prevent omissions in response. Specifically, this Database Management Department is built upon a distributed storage system supporting petabyte-scale data. It automatically creates copies on multiple physical nodes when writing data. It completely prevents data loss in the event of hardware failure. In the background, it continuously performs indexing processing on the accumulated text data using a full-text search engine. It constructs an inverted index structure. This database management department provides the ability to retrieve relevant data from millions of past question-and-answer and response records with millisecond-level response speeds. This database management department implements data lifecycle management strategies. This database management department automatically archives old data that has not been accessed for a certain period to inexpensive cold storage. This database management department optimizes storage costs. This database management department assigns metadata to each data record, including the creator, creation date and time, update history, and access permissions. This database management department enforces authorization processing based on strict access control lists. This database management department physically and logically blocks unauthorized access to highly confidential legal data. This database management department guarantees the reliability, availability, and confidentiality of data throughout the system. This database management department plays an indispensable technical role in maintaining a high-quality knowledge base for enabling generative AI to perform accurate reasoning.

[0059] The Generative AI Execution Department, based on prompts and data received from the Legal Amendment Information Acquisition Department, Affected Department Determination Department, and Response Plan Generation Department, performs natural language processing and knowledge reasoning to generate various outputs such as summaries, extractions, analyses, suggestions, document generation, FAQ generation, script generation, and Q&A. For example, it performs various generation tasks such as "extracting key points of legal amendment provisions," "summarizing practical explanations of guidelines," "analyzing the correlation between business content and legal amendment content," "automatically generating change application plans," "suggesting new business opportunities," "automatically creating briefing scripts," "automatically generating FAQs," and "chatbot-based Q&A." Input prompts to the generative AI include, for example, "Summarize the following legal amendment-related documents and extract the main amendment points, implementation date, additional obligations, penalties, and transitional measures," "Check the following departmental business content against the key points of legal amendments and list the affected departments and necessary response plans," "Create a briefing script and FAQ based on the following content," and "Answer the following questions by combining the key points of legal amendments and departmental business content." Specifically, this generative AI execution unit is built around a large-scale language model with billions to hundreds of billions of parameters. This large-scale language model employs a Transformer architecture with a self-attention mechanism. The generative AI execution unit uses a tokenizer to convert the input text data into a fixed-length sequence of lexical units. This sequence of lexical units is then mapped to a high-dimensional embedding vector space. For example, the generative AI execution unit represents the input data as a dense vector of 512 to 4096 dimensions. The generative AI execution unit receives text data of legal amendments, business manuals from various departments, and past response records as input tensors. A multi-layer self-attention network is applied to the input tensors. The generative AI execution unit calculates contextual dependencies. The generative AI execution unit weights the importance of each lexical unit. The generative AI execution unit generates a summary text, a list of extracted keywords, or structured data of response solutions as output. For example, this generative AI execution unit outputs a date string representing the implementation date, itemized text representing additional obligations, and numerical data representing the penalty content as output data. This generative AI execution unit then sends the output text data to the subsequent response generation unit and user interface unit. Based on the received text data, the response generation unit automatically formats the internally prescribed amendments. Through this structure, the system eliminates the existing process of manual document reading and understanding by humans. This system utilizes vector operations in high-dimensional space to quickly and accurately determine the semantic relationships between specific documents. This system significantly improves the information processing speed in the initial stages of responding to legal amendments. In fields requiring strict compliance, such as financial institutions and medical institutions, this system effectively minimizes the risk of business disruption during legal amendments.This large-scale language model is pre-trained using a large amount of legal and business practice documents. It is further fine-tuned to suit the specific business domain of each enterprise. In this fine-tuning process, the system uses the cross-entropy error function to optimize the model's weights. Gradient descent is used to update the parameters. Hint engineering techniques are employed to assign specific instructions to the input text. Zero-shot or few-shot learning methods are used to generate appropriate outputs even for unknown legal modifications.

[0060] The Log Management Department records the entire processing history of the system, user operation history, and AI output history, which is used for future auditing and to ensure traceability. For example, it records in detail who uploaded which legal amendment document and when, which department reviewed and edited which response plan, what AI answer was given to which question, and who received which notification. It also manages the progress of legal amendment responses and the retrieval history of unresolved tasks. Furthermore, it records user feedback and corrections to the AI's output, utilizing this data to improve AI accuracy. Finally, it records collaboration history with external clients and group companies as evidence of information sharing. Specifically, the Log Management Department uses a distributed database system and an append-only data structure. All events occurring within the system are retrieved as timestamped event logs. The retrieved event logs are encrypted using a hash function. The encrypted event logs are stored in blockchain technology or tamper-proof storage. The Log Management Department generates structured data containing user identifiers, data identifiers of the objects being operated on, the type of operation, and the date and time of the operation. This log management department saves complete pairs of input prompts and output text from a large-scale language model as log data. It extracts discrepancy data from user corrections to the output of the large-scale language model. This discrepancy data is then formatted into teacher data for the natural language processing model's relearning. The formatted teacher data is then transferred to the learning data repository. Through this structure, the system transcends simple history storage, building a continuous self-improvement cycle for the artificial intelligence model. The system models user operation history as a multi-dimensional vector. It uses anomaly detection algorithms to detect unauthorized access and misoperations that deviate from normal operating patterns in real time. Based on the detection results, the system generates immediate alerts to system administrators. The system mathematically guarantees the integrity of audit evidence. It fully automates information security auditing and reporting processes to regulatory authorities. This system delivers a significant technological boost to the transparency of compliance management across large-scale enterprise groups and the entire supply chain. The log management department regularly compresses log data and archives it to cold storage. It optimizes storage capacity consumption. It dynamically builds retrieval indexes, achieving millisecond-level retrieval responses for petabyte-scale log data.

[0061] The Notification Management Department automatically distributes notifications to various terminals, including legal amendment information and response plans, updated data, briefing guides, and FAQ updates. For example, it can instantly receive notifications such as "New legal amendment information added," "This department has been selected as an affected department," "New response plan generated," "Briefing guide," and "FAQ updated." It also automatically distributes reminder notifications to departments with unfinished tasks. Furthermore, it can uniformly distribute notifications to external clients and the group company. It also manages read / unread status of notifications and references notification history. Specifically, the Notification Management Department adopts an event-driven messaging architecture. It uses a publish-subscribe model to perform asynchronous message distribution. The department receives event triggers indicating status changes from various modules within the system. Based on the received event triggers, it dynamically generates notification payloads. It retrieves user attribute information, department affiliation, and current task load from the configuration file database. Finally, it uses machine learning-based recommendation algorithms to determine the optimal notification timing and channel for each user. This notification management department, for example, sends high-urgency legal amendment alerts as push notifications to mobile terminals. It sends low-urgency periodic reports as emails. The notification payload includes identifiers of the legal amendment data to be addressed and deep links to the corresponding tasks. When a user clicks on a notification, the system directly launches the corresponding task screen. Through this structure, the system reduces the cognitive load on users caused by information overload. The system physically prevents the omission of important compliance tasks. The system effectively optimizes network bandwidth consumption and reduces the overall communication load. The notification management department tracks the distribution status, arrival confirmation, and read status of each notification in real time. If a notification remains unread for a certain period, it automatically generates a secondary notification for higher-level management according to escalation rules. The notification management department integrates with a multilingual translation module to distribute notification messages translated into local languages ​​to overseas branches of global enterprises in real time. This system enables unified management of global legal compliance within multinational corporations.

[0062] The Historical Records Management Department manages past legal amendment responses and departmental responses in a chronological order, utilizing this information as a reference for future legal amendments. For example, it records, in chronological order, which department responded to past legal amendments, what documents were revised, what questions were asked most frequently, and what new business proposals were made. It also accumulates past briefing materials, FAQs, and Q&A records for reuse. Furthermore, it manages past response progress and reminder records, using them for analysis of potential omissions. It also records responses from external clients and group companies, helping to improve the group's overall legal amendment response capabilities. Specifically, the Historical Records Management Department combines a chronological database and a graph database. It defines past legal amendment events, responding departments, revised documents, and related Q&As as nodes. It constructs a knowledge graph representing the relationships between these nodes as edges. When new legal amendment information is input, the department uses a graph neural network to perform similarity searches on the knowledge graph. The department also identifies similar past legal amendment events. This historical record management department extracts sequences of response tasks performed in specific past events. This department provides these extracted task sequences as templates for current legal amendment responses to the response plan generation department. This department inputs patterns of delays and errors that occurred in past responses into a machine learning model. This department predicts bottlenecks in future response processes. Based on the predictions, this department generates optimal resource allocation plans. Through this structure, the system persists the organization's tacit knowledge as an explicit data structure. This system enables highly predictive responses based on past experience. This system delivers the technological effect of repeatedly reducing the human resources and time required for legal amendment responses. This department uses natural language processing technology to cluster frequently occurring topics from past question-and-answer logs. Based on the clustering results, this department automatically updates a general FAQ database. This department calculates quantitative scores for the response performance of external clients and group companies. This department outputs this score as a risk assessment indicator for the entire supply chain. This system strongly supports compliance integration processes during corporate acquisitions and organizational restructuring.

[0063] The User Interface (UI) department provides access to the server from terminals for responsible personnel in various departments, enabling them to view and download generated summaries, lists of affected departments, response plans, notification materials, FAQs, briefing scripts, etc. For example, legal personnel can confirm "what matters should our department address in this revision," "which materials should be revised," and "what explanations should be given at the briefing." It also provides data retrieval and filtering functions, as well as customized displays for each department. Furthermore, it allows for feedback input on AI-generated data and submission of revision plans. Dedicated screens for external clients and the group company are also available. Specifically, this UI department employs a single-page application architecture, implemented as a dynamic web interface. It communicates asynchronously with the server-side application programming interface. Based on user authentication information and access permission levels, it dynamically renders the data components to be displayed. It structurally configures text data generated by large-scale language models on a visual dashboard. It visualizes the list of affected departments as a network diagram. It depicts the progress of response plans in Gantt chart form. It receives search queries from users. This user interface department transmits the query to a semantic search engine using natural language processing. This user interface department not only suggests results with perfectly matching keywords but also those with semantically similar keywords. This user interface department collects user feedback on AI-generated materials in the form of star ratings or text reviews. This user interface department serializes the collected feedback data and sends it to the log management department. Through this structure, the system provides complex legal amendment information in an intuitive and understandable form. This system minimizes the number of user steps, dramatically improving business efficiency. This system leverages technology to optimize the rendering load on client terminals, ensuring comfortable operability even from low-spec devices. This user interface department employs a responsive design, automatically adapting to access from smartphones and tablets. This user interface department provides a secure sandbox environment for external clients, allowing access only to specific data domains. This system achieves seamless collaboration with external partners while completely eliminating the risk of information leakage.

[0064] The Document Viewing and Editing Department provides both direct access to generated materials and editing capabilities for additions, deletions, and modifications by department. For example, the Information Management Department can add department-specific applications to the AI-generated application manual revision plan. The Legal Department can add company-specific clauses to the AI-generated contract templates. The General Affairs Department can insert company-specific examples into AI-generated educational materials. The External Affairs Department can add the latest administrative guidance examples to AI-generated external explanatory materials. Specifically, this Document Viewing and Editing Department is implemented as a rich text editor component with real-time collaborative editing capabilities. It employs operation conversion algorithms or conflict-free copy data types. It mathematically resolves conflicts arising from simultaneous editing operations by multiple users on the same document. It expands the initial document generated by a large-scale language model in memory as a baseline. It captures keystrokes and text insertion / deletion events by users in milliseconds. It asynchronously sends edited discrepancies as difference patches to the server. This document review and editing department uses different metadata tags to distinguish between AI-generated suggested text and manually entered text by humans. It integrates with the version control system. This department supports rollback to any past state of a document. It provides a function that triggers a regeneration request to a large-scale language model via a context menu when a user selects a specific paragraph. Through this structure, the system seamlessly integrates the generative capabilities of AI with human expertise. This system dramatically reduces the delivery time required for document production. This system provides the technical benefits of ensuring full traceability of editorial history and strengthening the quality assurance process for compliant documents. This department inputs the text of unique clauses added by the legal department into the natural language processing model. This department verifies in real time whether these clauses contradict existing laws and company regulations. When a contradiction is detected, the department displays a warning highlight on the screen. This system proactively prevents human error in the contract and manual production process.

[0065] In the chatbot response department, after briefings and during daily operations, when a user inputs a question from their terminal, the content is sent to the server. The generative AI on the server, referencing the question content, key points of legal amendments, departmental business content, and past Q&A history, generates the best answer and sends it back to the terminal. For example, for specific questions such as "Under what circumstances do the new record-keeping obligations apply?", "Where should the existing user manual be revised?", "At what time should notifications be sent to external clients?", and "Which part of the educational materials should be updated?", the AI ​​provides an immediate answer. As an example of a prompt to the generative AI, one could say, "For the following questions, provide a practical and specific answer, combining key points of legal amendments and departmental business content." Specifically, this chatbot response department is structured as a conversational AI system employing a retrieval-enhanced generative architecture. This chatbot response department receives natural language question text input from the user's terminal. The chatbot response department inputs this question text into an embedding model, converting it into a high-dimensional question vector. For example, this chatbot response department represents the question text as a 1024-dimensional floating-point vector. This chatbot response unit performs an approximate nearest neighbor search on the vector database. It extracts text blocks with high cosine similarity to the question vector, including key legal amendments, departmental business manuals, and past question-and-answer history. The unit incorporates these extracted text blocks as contextual information into the question text. It then formats the combined text as input prompts for a large-scale language model. The large-scale language model performs inference processing. It receives the response text output from the large-scale language model. It automatically embeds hyperlinks to reference source documents within the response text. Finally, it sends the final response data to the user terminal. This structure suppresses the illusion phenomenon inherent in large-scale language models. The system consistently generates accurate answers based on the latest and internally relevant facts. It significantly reduces the number of standard consultations to the legal department, concentrating professional resources on highly analytical legal matters. For follow-up questions from users, the system maintains the dialogue history in a memory buffer. It enables multi-turn dialogues that maintain context. The chatbot's response unit calculates a confidence score for each response. If the confidence score falls below a predetermined threshold, the response unit automatically escalates the chat session to a human legal representative.

[0066] The notification receiving department receives notifications from the server in real time and distributes them via pop-ups, emails, chat messages, and other means. For example, it can instantly receive notifications such as "New legal amendment information added," "This department has been selected as an affected department," "New response plan generated," "Guide to holding briefings," and "FAQ updated." It also automatically distributes reminder notifications to departments with unfinished tasks. Furthermore, it can manage read / unread status of notifications and reference notification history. It can also uniformly distribute notifications to external clients and the group company. Specifically, this notification receiving department is implemented as a persistent background process or service worker thread on the client terminal. It uses the WebSocket protocol or server-sent events to establish a continuous two-way communication channel with the server. It asynchronously listens for notification payloads sent from the server. Upon receiving a notification payload, it invokes the operating system's native notification application programming interface. It displays a visual pop-up alert on the screen. It parses the priority flags contained in the notification payload. If the priority is high, it activates an audible alarm and vibration function. This notification receiving unit detects when a user clicks on a notification. Using this event as a trigger, it routes the notification to the corresponding task screen within the system. The unit records the completion of the notification display and the user's click action along with a timestamp. This status data is then sent back to the notification management unit on the server. This configuration ensures reliable message delivery even in environments prone to network latency and disconnections. The system does not interrupt the user's work context, delivering important compliance information promptly. This system maximizes the initial speed of responding to legal changes across the entire organization. The unit maintains a cache database of notification history in local storage. It enables the retrieval and viewing of past notification content even offline. The unit includes a Webhook integration module for connecting with third-party business chat tools. This system maximizes the utilization of the enterprise's existing communication infrastructure.

[0067] The Operation Log Sending Department sends the operation history from the terminal to the server for unified management of system utilization and response progress. For example, it records in detail which responsible person viewed and edited which document, received which notification, input which question into the AI, and approved or rejected which response plan. For example, it automatically calculates response progress based on the operation history and identifies departments and tasks that have not yet been addressed. For example, it also manages the operation history of external clients and the group company in a unified manner, visualizing the response status of the entire group. For example, it can also analyze system utilization and evaluate the effectiveness of AI suggestions based on the operation logs. Specifically, this Operation Log Sending Department functions as an event tracking module assembled in the user interface layer of the client terminal. This Operation Log Sending Department captures user interaction events such as mouse clicks, keystrokes, scrolling, screen dwell time, and focus shifts from the Document Object Model. This Operation Log Sending Department converts the captured raw data into structured log objects containing user identifiers, session identifiers, timestamps, and element identifiers of the operation objects. To suppress network bandwidth consumption, this Operation Log Sending Department temporarily accumulates multiple log objects in a memory buffer. This operation log sending unit sends logs to the server in batches at specified time intervals or when the buffer capacity threshold is reached. In the event of a network error during transmission, the unit automatically performs retransmission using an exponential backoff algorithm. Through this configuration, the system reliably collects high-resolution operation data without degrading client terminal performance. The system uses the collected operation data to cluster user behavior patterns in a multi-dimensional space. The system quantitatively calculates the acceptance rate and correction frequency of user responses to AI-generated solutions. The system leverages the technology of early detection of deterioration in the inference accuracy of AI models and automatically generates relearning triggers. From a privacy perspective, the operation log sending unit hashes or masks personally identifiable information and highly confidential input text on the client side. The unit only sends anonymized data to the server. This system, while adhering to strict data protection regulations, enables in-depth analysis of the overall system's utilization.

[0068] As described above, the system described in this embodiment achieves a unified and efficient automated response to the entire process of legal amendments through inter-departmental collaboration. Furthermore, by effectively utilizing generative AI, it comprehensively and specifically supports all business operations, including the collection and analysis of legal amendment information, identification of affected departments, generation of response plans, document preparation, briefing operation, Q&A, progress management, notification sending, historical record management, and external collaboration. Specifically, this system employs a distributed computing architecture with multiple independent microservices loosely coupled via application programming interfaces (APIs). The system uses containerization technology to encapsulate each processing module. The system dynamically scales each module on a container orchestration platform. The system models the data flow from the input of legal amendment information to the final response completion as a directed acyclic graph. The system configures an AI pipeline centered on a large-scale language model at each node. The system comprehensively executes natural language processing, knowledge graph reasoning, and machine learning-based prediction algorithms. The system converts input unstructured text data into machine-readable structured data. Through operations in a high-dimensional vector space, the system automatically resolves the complex semantic mapping between laws and enterprise business. This architecture fundamentally restructures existing compliance operations, which rely on the limits of human cognitive abilities, using computer science methods. It eliminates information processing bottlenecks, dramatically reducing the total time required to respond to legal changes from months to days. By eliminating the possibility of human error, it mathematically improves the accuracy of compliance across the entire enterprise. It flexibly utilizes the computing resources of cloud computing environments. Even under conditions of simultaneous, large-scale legal changes, it maintains consistent performance, handling massive tasks in real-time through parallel distributed processing. This system provides a robust technological foundation for balancing sustainable business growth with minimizing legal risks.

[0069] (Implementation steps) Step 1: Collection and Input of Legal Amendment Information. Legal, general affairs, information management, and foreign affairs personnel upload legal amendment documents, guidelines, public opinions, legislative discussion records, executive branch interpretations, industry group opinions, precedents, administrative guidance documents, and various news reports from the terminal. Additionally, information can be automatically retrieved from external legal databases or government websites via API linkage or web scraping. For example, documents in various formats, such as PDF of amended legal provisions, HTML of guidelines, and Word documents of industry group opinions, are considered. Specifically, this system initiates a multimodal data acquisition pipeline. The system receives binary data streams sent from user terminals via secure communication protocols. The system parses the received file extensions and magic numbers, automatically identifying the file format. For portable document format files, the system applies an optical character recognition engine and layout parsing algorithms. The system accurately extracts string data from graphical text or complex table structures. For Hypertext Markup Language (HTML) files, the system parses the Document Object Model (DOM) tree. The system filters out unnecessary decorative tags and navigation elements, extracting only pure text content. This system periodically sends representative status transmission application programming interface (API) requests to external legal databases. It automatically acquires the latest legal amendment data. It uses a web crawler module to traverse specific administrative agency domains. It detects updated web pages and extracts differential data. Through this structure, the system normalizes scattered unstructured data in various formats into a unified plain text format suitable for subsequent artificial intelligence processing. The system completely eliminates manual human intervention in the data collection process. This system achieves a significant technological leap in improving the comprehensiveness and immediacy of information. The system performs preprocessing on the extracted text data, including character encoding unification, noise character removal, and word normalization using lexical analysis. The system stores the preprocessed text data along with metadata in a distributed object storage. The system guarantees high-speed read access to large-scale datasets.

[0070] Step 2: Parsing and Summarizing Legal Amendment Information (Utilizing Generative AI). The legal amendment information acquisition unit on the server transmits uploaded or automatically acquired documents to the generative AI execution unit for parsing using natural language processing technology. Input prompts to the generative AI include, for example, "Summarize the following legal amendment provisions and extract the main amendment points, implementation date, added obligations, deleted provisions, strengthening of penalty provisions, and transitional measures.", "From the following full text of the guidelines, summarize the Executive Yuan's interpretation policy, practical considerations, and impact on the industry.", and "From the following public opinion results, summarize the main opinions and the Executive Yuan's response." The generative AI automatically extracts and summarizes the main points of the amended law, amendment points, implementation period, new rules to be followed, and impact on the industry, and stores this information in the server's database management unit. Specifically, this system tokenizes the preprocessed text data and, in accordance with the input context window limitations of the large-scale language model, divides it into appropriately sized blocks. The system provides system prompts for each block, including task-specific instructions. For example, this system can provide a text array of legal amendment provisions lasting tens of thousands of words as input data to the large-scale language model. This system uses a self-attention mechanism within a Transformer network of a large-scale language model to compute long-distance dependencies in text. It evaluates the contextual semantics of legal terminology or complex syntax in a high-dimensional vector space. As output from the large-scale language model, it generates structured data. Output data includes, for example, strings representing the effective date "April 1, 2024," text representing additional obligations such as "record production at the time of third-party provision," and numerical data representing penalties such as "fines of less than one million yen." For each extracted element, the system calculates a confidence score. Data with confidence scores exceeding a threshold are mapped to relational tables in the database management department and stored. Through this structure, the system can complete the reading comprehension and key point organization of complex legal documents, which would take humans several days, in seconds to minutes. It eliminates omissions in information extraction or biases in interpretation, providing homogeneous and highly accurate parsing results. This system provides high-quality structured data for generating subsequent department-specific or response plans. It integrates multiple different large-scale language models. This system integrates the outputs of various models using majority voting or cross-validation. It strongly prevents the infiltration of erroneous information due to the illusion of a single model. The extracted summary data is vectorized and indexed in a vector database. This system also speeds up subsequent similarity retrieval processing.

[0071] Step 3: Specific Impact on Departments (Utilizing Generative AI). The Department Impact Assessment Department will compare the business content, business processes, areas of responsibility, past legal amendment response history, organizational charts, business process diagrams, lists of managed assets, lists of external clients, overviews of relevant systems, data categories used, various regulations and manuals, educational materials, past internal company notices, and past internal audit findings of each department registered on the server with the key points of legal amendments extracted by the Legal Amendment Information Acquisition Department, and use generative AI to analyze the correlation. The generative AI will be prompted with messages such as, "Compare the following key points of legal amendments with the business content databases of each department, list the affected departments, and extract highly correlated business processes and managed assets." The AI ​​will automatically determine, for example, that the Information Management Department, responsible for "the management and use of personal information," is an affected department by the amendment to the Personal Information Protection Law. Specifically, this system will input the extracted text data of the key points of legal amendments into the embedding model to generate legal amendment feature vectors. In parallel, this system will also vectorize the text data of each department's business content, managed assets, and various manuals to generate a set of department configuration file vectors. This system calculates cosine similarity or inner product between the feature vector of legal amendments and the configuration file vectors of each department. Based on distance in the vector space, it quantitatively assesses the strength of the semantic correlation between legal amendments and the business operations of each department. Departments with similarity scores exceeding a predetermined threshold are extracted as candidates. The system combines the configuration file data of the extracted candidate departments with the key points of the legal amendments to construct input prompts for a large-scale language model. This system enables the large-scale language model to perform impact assessment based on logical reasoning, rather than simple keyword matching. As output from the large-scale language model, the system obtains structured data containing identifiers of affected departments, names of specific affected business processes, and risk scores indicating the degree of impact. For example, the system generates a tag set as output data such as "Information Management Department," "Customer Database Application Process," and "High Risk." Through this structure, unlike existing retrieval systems that rely on surface-level word matching, this system achieves a high-level matching that deeply understands the actual business situation and the essence of legal regulations. This system completely prevents specific omissions of the scope of impact of legal amendments. This system leverages the technical effects of automating the assessment of compliance risks across the company and prioritizing optimized responses. The system maps the list of specific affected departments as nodes on a graph database. This system tracks the business dependencies between departments as edges. It employs network analysis algorithms that not only target directly affected departments but also indirectly affected related departments.

[0072] Step 4: Automatic Generation of Response Plans and Materials (Utilizing Generative AI). The Response Plan Generation Department uses generative AI to automatically generate specific response plans based on legal amendments for each department identified as affected. Input prompts into the generative AI include: "Based on the following key points of the legal amendments and the department's business content, create necessary application change plans, data revision plans, revision plans for existing manuals, and new business proposals." and "Create a briefing script and FAQ draft based on the following content." The AI ​​automatically generates application change plans and data revision plans, new business proposals, briefing scripts, and FAQ drafts, such as "Reassess the application process for third-party provision of personal information records," "Add new contract terms," ​​"Revise the shareholders' meeting operation manual," and "Revise the communication service usage regulations." Specifically, this system integrates configuration file data of the specific affected departments, key points of the legal amendments, and past similar response experience data to dynamically generate context-rich input prompts. This system inputs these prompts into a large-scale language model. The system then performs an autoregressive label generation process within the decoder network of the large-scale language model. This system uses a cluster search algorithm or kernel sampling method to generate the most appropriate and fluent text sequence in the context. As output from a large-scale language model, the system generates a list of specific action items, differential text of internal company regulations to be amended, and presentation scripts for briefings. For example, as output data, the system outputs specific amendment instructions and text blocks of amendment schemes for phrases such as "add the following to Article 3 of the existing privacy policy." The system converts the generated text data into structured document formats such as Markdown or Hypertext Markup Language. The system registers the generated response schemes as new records in the task management database. Through this structure, the system automatically translates the abstract external factor of legal modification into specific business tasks that departments can immediately execute. The system reduces the enormous manual time required to create materials from scratch. The system delivers a dramatic technical effect by shortening the delivery time from the initiation of a response strategy to its implementation. The system applies a rule-based checker module to the generated response schemes to verify compliance with internal compliance benchmarks or brand guidelines. In cases containing inappropriate expressions or contradictions, the system automatically iterates through a feedback loop containing amendment instructions to the large-scale language model. This system autonomously guarantees the quality of the final output within the system itself.

[0073] Step 5: Sending, viewing, and editing generated materials and response plans. Through the user interface department and the material viewing and editing department, responsible personnel from each department can access the server from their terminals to view and download generated summaries, lists of affected departments, response plans, notification materials, FAQs, briefing scripts, etc. Responsible personnel can either directly use AI to generate materials or supplement and modify them by department. For example, the Information Management Department can add records of its own applications to the revised application manual generated by AI, and the Legal Department can add its own clauses to contract templates. Specifically, this system configures various generated materials on a distributed file system, assigning each material a unique resource identifier. This system evaluates user authentication tokens and role-based access control lists. This system establishes a secure data transmission channel that only allows access to materials to users with confirmed permissions. This system provides a high-speed rendering engine using a virtual document object model to the web browser on the client terminal. This system uses different color codes or metadata tags to visually distinguish AI-generated text from manually edited text. This system captures change events every time a user changes text in the editor. This system sends changes as differential patches to the server via asynchronous communication. The system applies these differential patches on the server side, maintaining the document's up-to-date state in memory. When multiple users are editing the same document simultaneously, the system uses an operation conversion algorithm to resolve editing conflicts in real time. Through this architecture, the system achieves a hybrid workflow that seamlessly integrates automated generation by artificial intelligence with professional human fine-tuning. The system eliminates the complexity of version control for documents. The system maximizes the collaborative efficiency of the entire organization. The system accumulates user editing history as teacher data for machine learning models. The system learns patterns in user corrections. In subsequent document generation, the system fine-tunes the parameters of the large-scale language model to output text that better reflects user intent or internal company habits. Through continuous utilization, the system self-regulates and improves the overall generation accuracy.

[0074] Step 6: Implementation of Briefings and Notification Activities. Departments utilize AI-generated briefing scripts and notification materials to conduct briefings, internal company training sessions, and briefings for external clients. For example, the General Affairs Department holds a briefing to explain the operational changes announced at the shareholders' meeting, and the Information Management Department conducts internal training on the new personal information management process. AI-generated FAQs and Q&A materials are also utilized. Specifically, this system has an export module that automatically converts automatically generated briefing scripts and notification materials into presentation software file formats or standard specifications for e-learning platforms. This system uses a speech synthesis AI model to automatically generate narration audio files composed of natural human voices from text data. This system synchronizes the generated audio files with the presentation slides, dynamically constructing on-demand multimedia training content. This system automatically uploads this content to the company's internal video server or learning management system via an application programming interface. This system collects employees' viewing and listening history, comprehension test scores, and content access logs. This system integrates the collected data into a central database. This system sends automatic reminder emails to the management department for employees who did not attend the presentation. This architecture dramatically reduces the physical preparation and conference costs required for in-house communication of legal changes. It provides instant, homogeneous, and high-quality educational opportunities to all employees. The system delivers near 100% attendance at compliance training sessions, improving legal literacy across the entire organization. It captures the audio of Q&A sessions from microphones. It uses a speech recognition model to transcribe the audio in real time. The system then inputs the transcribed questions into a large-scale language model, instantly generating draft answers and displaying them on the presenter's teleprompter. While powerfully supporting briefings, the system is self-expanding, automatically adding new Q&A data to the FAQ database.

[0075] Step 7: Q&A via Chatbot (Utilizing Generative AI). Through the chatbot response department, after briefings or in daily operations, when a user enters a question from their terminal, the content is sent to the server. The generative AI on the server, referencing the question content, key points of legal amendments, departmental business content, and past Q&A history, generates the best answer and sends it back to the terminal. Examples of prompts for the generative AI could include, "For the following questions, provide practical and specific answers based on key points of legal amendments and departmental business content." For instance, immediate answers could be given to questions such as "Under what circumstances do the new record-keeping obligations apply?" or "Where should the existing user manual be revised?" Specifically, this system receives natural language query strings input from the user's terminal chat interface. The system performs lexical analysis and stop word removal on the query string. The preprocessed query is then input into a Transformer-based encoder model to generate a dense vector representation. The system applies an approximate nearest neighbor search algorithm to the vector groups of key points of legal amendments, business manuals, and past FAQs stored in the vector database. This system rapidly retrieves and extracts the top K text blocks with the highest similarity to the query vector. The extracted text blocks are appended as context to the query string, dynamically constructing input hints for the large-scale language model. The system enables the large-scale language model to perform inference. As output from the large-scale language model, the system generates structured response data containing direct answers to the question, internal company regulations as the basis, and hyperlinks to relevant manuals. For example, the system generates the text "New record obligations apply in all cases of providing personal data to third parties. See the link to the 5th edition of the user manual for details." Through this structure, the system completely overcomes the obsolescence or illusion of knowledge inherent in the large-scale language model by using retrieval-enhanced generation techniques. The system consistently provides accurate answers based on the latest internal company rules with millisecond-level latency. The system significantly reduces consultation traffic to the legal department, minimizing business downtime. The system collects binary feedback from users, indicating whether the information is "helpful" or "not helpful." The system uses reinforcement learning algorithms to continuously optimize the weights of the retrieval model or the generation parameters of the large-scale language model.

[0076] Step 8: Progress, History, and Notification Management. The Log Management Department, Notification Management Department, and History Management Department work together to unify the management of the entire processing history, user operation history, AI output history, response progress, notification history, and past legal amendment response history. The Notification Management Department automatically sends legal amendment information or response plans, data updates, briefing guides, FAQ updates, and reminders for unresolved tasks to all terminals. The History Management Department records the response content or progress status of past legal amendments in chronological order and uses it as reference information for future legal amendments. Specifically, this system uses a distributed message queue system to aggregate event logs sent asynchronously from various modules. This system uses a stream processing engine to perform real-time parsing, filtering, and centralized processing of log data. This system writes the processed data in parallel to a time-series database and a data warehouse. This system manages the status of each department's response tasks using a state transition model. This system compares the task completion deadline with the current status. This system automatically detects high-risk tasks that have exceeded their deadlines. Based on the detection results, this system triggers the Notification Management Department. This system dynamically generates and sends escalation-level alerts to the relevant responsible parties and their supervisors. It archives complete project data for all legal amendment responses within the historical records management department. The system constructs a multi-dimensional data cube from project initiation to completion, identified challenges, and the frequency of AI utilization. This structure allows for real-time visualization of the entire organization's compliance response status on a bird's-eye view dashboard. The system enables early detection of delays or bottlenecks, facilitating proactive intervention. It delivers the technological advantage of minimizing project management costs. The system applies machine learning-based regression analysis to accumulated historical data. It accurately predicts the human resources and time required for similar legal amendment events in the future. This system strongly supports data-driven business decision-making.

[0077] (specific use case) For example, consider a scenario where laws concerning personal information protection are amended, introducing new obligations for businesses regarding record-keeping when providing information to third parties, reporting obligations upon leakage, changes to the definition of anonymized information, and strengthened regulations on cross-border transfers. In this case, legal personnel upload amended legal provisions, guidelines, public opinions, interpretations from the Executive Yuan, insights from industry groups, relevant case law, administrative guidance documents, and related news materials from the terminal to the system, or automatically retrieve them from external legal databases or administrative agency websites via API linkage or web scraping. The Legal Amendment Information Acquisition Department receives these documents in various formats and generates prompts for the Generative AI Execution Department, such as "Summarize the following amended legal provisions and extract the main amendments, implementation dates, added obligations, deleted provisions, strengthened penalties, and transitional measures," or "Summarize the Executive Yuan's interpretation policy, practical considerations, and impact on the industry from the following full-text guidelines," instructing the system to analyze them. Specifically, in the specific use case of amending the Personal Information Protection Law, this system enables the operation of an advanced natural language processing pipeline. This system aggregates hundreds of pages of amended legal provisions in PDF format sent from the terminals of legal professionals, as well as data obtained from administrative agencies' servers via application programming interfaces, into a central storage. Using optical character recognition (OCR) and a document parser, the system converts this unstructured data into arrays of plain text. The system then divides these text arrays into blocks of thousands of lexical units. Each block is formatted as an input tensor for a large-scale language model. The system provides the large-scale language model with specific system hints for understanding the legal context. The system drives the large-scale language model's inference engine to extract the core elements of the legal amendments from the input text data. As output data, the system generates, in the form of structured key-value pairs, names of new obligations such as "obligation to record information when providing to third parties," compliance requirements such as "obligation to report leaks," and constraints on international data transfers such as "strengthened regulations on cross-border transfers." Through this structure, the system completely replaces the existing mental process of reading massive legal documents and manually extracting and organizing relevant information. The system rapidly extracts the semantic relationships hidden within specific documents through vector operations in high-dimensional space. This system enables a precise grasp of the full picture of legal amendments within minutes. It significantly reduces the time required for initial analysis by legal departments and supports the rapid development of response strategies. For each extracted element, the system maintains pointers to the relevant original provisions or guidelines as metadata. Throughout subsequent processes, the system ensures that users can always easily access the traceability of information's primary source.

[0078] Generative AI automatically extracts and summarizes the main points, key modifications, implementation period, new rules to be followed, and impact on the industry of legal amendments. It stores key information such as "strengthened record-keeping obligations when providing information to third parties," "added reporting obligations in case of leaks," "changed the definition of anonymously processed information," and "new regulations regarding cross-border transfers" in the database management department. The affected department determines the impact by comparing the business content, business processes, areas of responsibility, asset lists, external client lists, related system overviews, data categories, regulations and manuals, educational materials, past internal company notices, and past internal audit findings of various departments (Information Management Department, Legal Department, General Affairs Department, Foreign Affairs Department, etc.) registered on the server with the key points of legal amendments extracted by the Legal Amendment Information Acquisition Department. It then inputs a prompt such as, "Compare the following key points of legal amendments with the business content databases of each department, list the affected departments, and extract highly relevant business processes and assets," into the generative AI. Specifically, this system normalizes the key points of legal amendments output from a large-scale language model and inserts them into a dedicated table in a relational database. This system uses a natural language processing model to vectorize stored key data, generating legal amendment feature vectors. It also reads configuration file data sets from various departments stored in the database management department in parallel. Similarly, it vectorizes text data such as "Customer Database Usage" from the Information Management Department, "Contract Review" from the Legal Department, and "Employee Data Management" from the General Affairs Department, generating departmental configuration file vectors. The system calculates cosine similarity in multidimensional space between the legal amendment feature vectors and the departmental configuration file vectors. It pairs departments with specific similarity scores exceeding a predetermined threshold with the key points of the legal amendments. Combining the data from these specific pairs, the system dynamically constructs input prompts for a large-scale language model. This system enables the large-scale language model to infer logical causal relationships between business processes and legal regulations, rather than simply keyword matching. As output from the large-scale language model, the system generates structured data representing a list of affected departments, specific affected business processes, and the directional direction of necessary responses. For example, as output data, the system generates mapping data showing that the "Information Management Department" meets the "reporting obligation in case of leakage" and needs to "revise the incident response process." Through this structure, the system highly integrates siloed business data within the organization with external legal and regulatory data. The system comprehensively and systematically assesses the impact of specific legal changes across the entire company. The system proactively prevents oversights in specific departments and coordination errors between related departments. The system delivers a mathematical and systematic reduction in overall enterprise compliance risk. The system visualizes the generated mapping data of affected departments as a network diagram through the user interface. The system provides decision support tools that enable management and legal personnel to intuitively grasp the company-wide impact and quickly determine the optimal allocation of resources.

[0079] Generative AI identifies departments directly affected by amendments to the Personal Information Protection Act, such as the Information Management Department (responsible for "Management and Use of Personal Information," "Database Management," "Management of External Clients," and "Access Control Management"). Similarly, the Legal Department (responsible for "Contract Management," "Specific Business Transactions," "Intellectual Property Management," and "Litigation Response") is identified as a department requiring revisions to contract templates and instruction manuals. The Response Generation Department inputs prompts into the AI, such as "Based on the following key points of the legal amendments and the department's business content, create necessary application change plans, data revision plans, revision plans for existing manuals, and new business proposals," or "Create a briefing script and FAQ draft based on the following content." For example, suggestions for the Information Management Department include "Reassessing the process for using records provided by third parties for personal information," "Developing new reporting procedures for leaks," "Revising contract templates with external clients," and "Revising access control regulations," as well as revisions to existing application manuals and internal company regulations. Suggestions for the Legal Department include "Adding new contract terms," ​​"Revising instruction manuals for specific business transactions," "Introducing new consent acquisition procedures," and "Reassessing litigation risk assessments." The General Affairs Department was given suggestions such as "updating employee training materials," "creating internal company notification documents," and "redesigning internal company training procedures." External Affairs Department was given suggestions such as "revising the communication service usage agreement," "revising explanatory materials for external negotiations," and "changing the reporting procedures to administrative agencies." Specifically, this system performs advanced reasoning processing using a large-scale language model. The Department Affected Assessment Unit inputs the text of the key points of the legal amendment and the business content text of each department into the Natural Language Processing (NLP) module. This NLP module performs lexicalization on the input text at the sub-word level. This module converts the lexical sequence into multi-dimensional word embedding vectors. This module inputs the converted vectors into a neural network with a self-attention mechanism. This network calculates the semantic correlation between the vector of the key points of the legal amendment and the vector of the business content. This network outputs an attention score representing the correlation. If the calculated attention score exceeds a predetermined threshold, the Department Affected Assessment Unit identifies that department as an affected department. The Response Generation Unit obtains the identifier of the identified affected department and the business content vector. The Response Generation Unit combines the vector of the key points of the legal amendment and the business content vector and inserts it into a prompt template. The solution generation department inputs the generated prompt string into a large-scale language model. The input data to this large-scale language model is, for example, a text tensor represented as a 512-dimensional floating-point array. This input data includes "Key Points of the Amendment to the Personal Information Protection Law" and "Business Content of the Information Management Department." The large-scale language model calculates the probability distribution of the next word's occurrence based on the input tensor. This model uses a beam search algorithm to explore the optimal word sequence.The output data from this model is an array of lexical IDs that constitute the generated response text. This output data is decoded into strings, for example, representing a specific action such as "reassessing the application process." The response generation unit formats the decoded text into structured data. This system sends this structured data to the database management unit. The database management unit stores the received data in a response table in a relational database. Through this processing, the system automatically generates comprehensive and accurate response solutions without relying on human intuition, through semantic parsing in a high-dimensional vector space. This configuration improves the comprehensiveness of responses to legal amendments. This configuration reduces the risk of compliance violations due to omissions in responses. This configuration enables efficient use of computer resources. This system is particularly effective in fields requiring strict compliance with laws and regulations, such as financial institutions and medical institutions.

[0080] Furthermore, regarding new business opportunities arising from legal amendments, the generative AI provides suggestions such as, "Based on the following key points of the legal amendments and the department's business content, propose new business solutions that are feasible due to the legal amendments." Suggestions include "launching new data utilization services," "developing consulting services to comply with the amended law," "developing services for new customer segments," and "entering new markets through business collaborations with other companies." The system also automatically generates informational materials, draft FAQs, briefing scripts, and time allocation plans for each department regarding the legal amendments. It uses prompts such as, "Create a briefing script and draft FAQ based on the following content," and creates scripts such as, "The first 10 minutes will be a summary of the amendments, the next 15 minutes will be an explanation of specific response strategies, and the remaining 5 minutes will be for Q&A," as well as FAQs covering topics such as "the scope of application of the new regulations," "the impact on existing business," "the specific steps for implementing the changes," "the necessity of internal company training," and "responses to external clients." Specifically, this solution generation department utilizes a finely tuned large-scale language model specialized for exploring new business opportunities. The response solution generation department generates an input tensor that combines a vector of key legal amendment points with a vector of the company's operating resources. This input tensor is, for example, a 124-dimensional floating-point array. This input tensor contains "specific details of the regulatory relaxation" and "a summary of the company's data." This large-scale language model processes the input tensor using multi-layered Transformer blocks. This model uses weight parameters pre-trained on a dataset of past successes for inference. The output data from this model is structured JSON-formatted text data containing the title, summary, and expected benefits of the new business proposal. This output data contains strings such as "consultation services for compliance with amended laws." The response solution generation department inputs the output new business proposals into the evaluation module. This evaluation module scores the feasibility and market size. This system only sends proposals that meet certain scoring criteria to the user interface department. The response solution generation department also performs the generation of informational materials and FAQs. The response solution generation department uses the key legal amendment points and response solution text as input to drive a summarization model. This summarization model performs an extractive summarization algorithm that extracts important sentences from long texts. This summarization model executes a generative summarization algorithm that reconstructs extracted sentences into natural-looking articles. The output data from this model consists of numerical data representing the schedule of a briefing session and the script text for each segment. This output data includes time allocation data such as "ten minutes" and "fifteen minutes." The system can also input the generated script text into a speech synthesis module to generate an automatic voice guide. Through this process, the system goes beyond simple legal compliance, creating business growth opportunities based on legal amendments. This structure significantly reduces the human resources and time required for new business planning. This structure enables logical business proposals based on objective data.This system functions as a tool to ensure a competitive advantage in the rapidly changing information and communication industry and fintech sector.

[0081] Each department's responsible personnel can view and download generated summaries, lists of affected departments, response plans, notification materials, FAQs, and briefing scripts through the user interface department and the document viewing and editing department, and can add, delete, and modify them as needed. AI-generated materials are also utilized for briefings, internal company training, and briefings for external clients. After briefings and in daily operations, when responsible personnel enter questions through the terminal, the chatbot response department generates optimal answers from the server's generative AI, prompting, "For the following questions, based on the key points of legal amendments and departmental business content, provide practical and specific answers." Questions such as "Under what circumstances do new record-keeping obligations apply?", "Where should the existing user manual be revised?", "When should notifications to external clients be made?", and "Which part of the educational materials should be updated?" are answered instantly. Specifically, the user interface department is implemented as a single-page application running on a web browser. The document viewing and editing department uses asynchronous communication protocols to send and receive data with the server. To enable multiple users to edit simultaneously, the document viewing and editing department implements exclusive control using operation conversion algorithms and differential synchronization processing. This chatbot response unit employs a retrieval-enhanced generation architecture. It receives the question text input from the user terminal. The response unit then inputs the question text into a natural language processing module, converting it into a question vector. This question vector is, for example, a dense vector of 768 dimensions. The response unit uses the question vector to retrieve a vector database. In this vector database, key points of legal amendments, company regulations, and past Q&A data are vectorized and stored. The response unit extracts the top five pieces of context text most similar to the question vector using cosine similarity calculation. The response unit combines the extracted context text and the question text to generate a prompt. The response unit inputs the generated prompt into a large-scale conversational language model. The input data to this model is a text tensor containing the question content and excerpts of relevant company regulations. The output data from this model is structured data containing the specific answer text to the question and the document ID of the reference source. This output data includes specific instructions, such as "Please revise Chapter 3, Section 2 of the User Manual." The response unit sends the output answer text to the user terminal and displays it on the screen. This system provides users with accurate and context-appropriate answers to their questions in real time, reflecting the company's internal situation. This significantly reduces the consultation load on legal and compliance departments. It also improves company-wide understanding and response speed to legal changes. This system is extremely useful in large enterprises with many employees or in manufacturing industries with complex business processes.

[0082] Furthermore, the Log Management Department, Notification Management Department, and Historical Record Management Department work together to unify the management of the entire processing history, user operation history, AI output history, response progress, notification history, and past legal amendment response history. The Notification Management Department automatically distributes notifications such as "New legal amendment information added," "This department has been selected as an affected department," "New response plan generated," "Guide to holding briefings," "FAQ updated," and reminders for unresolved tasks. The Historical Record Management Department records the response content and progress of past legal amendments in chronological order, which can be used as reference information for future legal amendments. Specifically, the Log Management Department uses a distributed tracing system to track all events within the system. The Log Management Department asynchronously collects log data output from various microservices via a message queue system. The Log Management Department stores the collected log data in a time-series database. This time-series database uses timestamps as the primary key to achieve high-speed writing and retrieval. The Notification Management Department adopts an event-driven architecture based on a publish-subscribe model. The Notification Management Department detects data update events in the database management department. The Notification Management Department evaluates predefined routing rules based on the type of update event. This notification management department generates a list of user IDs for notification recipients based on routing rules. This department then sends push notifications, emails, or messages to the user ID list via the company's internal chat tool. For each past legal amendment response project, this history management department stores all relevant documents, task completion status, and communication history in a graph database. This graph database represents documents and tasks as nodes and their dependencies as edges. When new legal amendment information is input, this history management department uses a graph neural network to retrieve similar past projects. This history management department extracts the response history of similar projects and provides it as contextual information to the response solution generation department. Through this process, the system automatically reuses past insights to achieve continuous improvement in the response process. This structure prevents the loss of tacit knowledge due to the transfer or departure of responsible personnel. This structure guarantees the reliability of the audit trail, reducing the response load during external audits. This system functions as a foundational system for strengthening governance in listed companies and financial institutions that require strict record keeping.

[0083] (Application Example 1) The specific processing flow corresponding to Use Case 1 will be described below. The various parts of the system described below are implemented by the data processing device 12 and the intelligent device 14. Furthermore, the data processing device 12 is referred to as the "server" and the intelligent device 14 is referred to as the "terminal".

[0084] (Implementation methods for carrying out the present invention) The embodiments of the present invention will be described in more detail using a system supporting responses to legal amendments in the field of electronic settlement services as an example. The system of the present invention consists of a server and a terminal. On the server side, there are a legal amendment information acquisition unit, an affected department determination unit, a response plan generation unit, a database management unit, a generative AI execution unit, a log management unit, a notification management unit, and a history management unit. On the terminal side, there are a user interface unit, a data viewing and editing unit, a chatbot response unit, a notification receiving unit, and an operation log sending unit. Specifically, this system adopts a scalable distributed processing architecture built on a cloud computing environment. The server is a computing cluster composed of multiple virtual machine instances. The server uses container orchestration tools to deploy each processing unit as an independent microservice. The legal amendment information acquisition unit, the affected department determination unit, and the response plan generation unit execute within independent containers. To accelerate large-scale matrix operations, the generative AI execution unit is configured on dedicated computing nodes equipped with multiple image processing semiconductor devices. To improve availability and fault tolerance, the database management unit is configured as a distributed database cluster spanning multiple availability zones. The terminal is an information processing device such as a personal computer, smartphone, or tablet. This terminal uses a secure, encrypted communication protocol to connect to this server. The user interface employs responsive programming techniques, updating the screen without delay in response to user actions. The operation log sending unit captures user actions such as mouse movements, clicks, and keyboard inputs. This unit aggregates the captured event data into batches of a certain size and sends them to this server as a background process. Through this configuration, the system provides high processing performance and stability for the massive transaction data and frequent legal amendments characteristic of the electronic settlement service field. This configuration ensures continued service delivery even if a part of the system fails. This system demonstrates extremely high reliability in mission-critical financial infrastructure requiring 24 / 7 / 365 uninterrupted operation.

[0085] The Legal Amendment Information Acquisition Department receives various legal amendment-related documents uploaded from terminals by legal personnel, risk management personnel, system developers, customer support personnel, and sales personnel. These documents include amendments to the Funds Settlement Act, the Act on Prevention of Transfer of Proceeds of Crime, Consumer Protection Act, the Electronic Bookkeeping Act, the Tax System Amendment Act, and the Personal Information Protection Act. The documents include amendments, guidelines, Executive Council notices, industry group opinions, precedents, administrative guidance documents, public opinions, and related news reports. Examples include PDF versions of Funds Settlement Act amendments, HTML guidelines, Word documents of industry group opinions, administrative notices automatically retrieved from the Executive Council website, summaries of public opinions, news clippings, explanatory reports from industry magazines, the Executive Council's FAQs, and past administrative guidance examples. Additionally, information can be automatically retrieved from external legal databases or government agency websites via API linkage or web scraping. For example, the department can automatically retrieve the latest legal amendment information daily from the Executive Council's API and periodically scrape guidelines and explanatory materials from industry group websites. The Legal Amendment Information Acquisition Department transmits received or acquired documents to the Generative AI Execution Department, generating prompts such as "Summarize the following amendments to the Payment Services Act, and extract the main amendments, implementation dates, additional obligations, strengthening of penalties, and transitional measures," "Summarize the Executive Yuan's interpretation policy, practical considerations, and impact on electronic settlement services from the following full text of the guidelines," "Summarize the main opinions and the Executive Yuan's responses from the following public opinion results," and "Extract the questions and answers anticipated on-site from the following Executive Yuan FAQs," and instructs the Generative AI to analyze them. Specifically, this Legal Amendment Information Acquisition Department executes a data pipeline that collects unstructured data from various data sources and converts it into structured data. This Legal Amendment Information Acquisition Department receives binary files uploaded from terminals. This Legal Amendment Information Acquisition Department uses a file format discrimination algorithm to parse the file extension and header information. For PDF and image files, this Legal Amendment Information Acquisition Department activates an optical character recognition engine to extract text data. This Legal Amendment Information Acquisition Department has a web crawler module. This web crawler module periodically sends HTTP requests based on a pre-defined list of URLs. This web crawler module parses the DOM tree from the acquired HTML document and extracts the main text. The legal amendment information acquisition unit performs authentication protocol processing for interaction with external APIs. This unit preprocesses the extracted text data. This preprocessing includes the removal of unwanted symbols, normalization, and stop word removal. The unit then segments the preprocessed text data into blocks. It combines the segmented text blocks with a prompt template containing extraction instructions. The generated prompts are then input into the generative AI execution unit. The input data to the generative AI execution unit is, for example, a text sequence consisting of thousands of tokens.This input data includes the "amendment proposal for Article 20 of the Payment Services Act" and the "implementation date of withdrawal." This generative AI execution unit parses the input text sequence, specifically matching the text span of the specified items. Through this processing, the system fully automates the massive manual document collection and reading tasks. This structure prevents information omissions, enabling the real-time incorporation of the latest legal regulations into the system. This structure improves the accuracy of data input, ensuring the quality of subsequent parsing and processing. In the fintech industry, where legal regulations frequently change, this system becomes an indispensable information collection foundation for maintaining compliance.

[0086] The Generative AI Implementation Department uses natural language processing technology to automatically extract and summarize the main points, amendments, implementation period, new rules to be followed, impact on the industry, additional obligations, strengthened penalties, and transitional measures of the revised law. Key points such as "strengthening of personal verification procedures," "addition of transaction record retention obligations," "new AML / CFT requirements," "changes to the upper limit of electronic settlement transactions," "new user explanation obligations," "addition of electronic signature requirements," "new identification of personal verification documents," "changes to suspicious transaction reporting standards," "stricter franchisee review standards," "new reporting obligations in case of system failures," and "introduction of a user compensation system" are stored in the Database Management Department. For example, it extracts from the Executive Yuan's notices "extension of the retention period for personal verification documents," "addition of new personal verification methods," and "changes to suspicious transaction reporting standards." It also extracts from industry group opinions "proposals to change the application process of industry standards," "the necessity of strengthening education for franchisees," "recommendations of new API specifications," and "strengthening of security requirements." Finally, it extracts from public opinions "concerns from users," "requests from businesses," "responses from the Executive Yuan," and "issues facing the industry as a whole." Specifically, this generative AI execution unit performs pipelined processing combining a named entity extraction model and a text summarization model. The generative AI execution unit feeds the input text block into a bidirectional Transformer encoder. This encoder calculates the context-dependent embedding vectors for each term. The generative AI execution unit feeds these embedding vectors into a Conditional Random Field layer to calculate the probability that each term belongs to a specific entity class. The generative AI execution unit predicts the label sequence with the highest probability and extracts the corresponding text span. The generative AI execution unit also feeds the extracted text span into an autoregressive decoder network. This decoder network converts the extracted information into a concise summary text. The output data from this generative AI execution unit is a JSON object containing key-value pairs of the extracted entities and the summary text. This output data includes, for example, the value "April 1, 2024" for the key "Implementation Period" and the value "Seven-year retention of transaction records" for the key "Additional Obligation". The generative AI execution unit sends the output JSON object to this database management unit. The database management department maps the received data to the columns of the legal amendment key points table and saves it. Through this processing, the system extracts meaningful structured data from unstructured text with high precision. This structure converts lengthy and complex legal documents into a computer-processable data format. This structure provides the foundational data for accurate logical reasoning in subsequent assessments of affected departments and the generation of response plans. In interpreting financial regulations with complex legal systems, this system achieves accuracy and overwhelming processing speed comparable to or exceeding that of human experts.

[0087] The Department of Impact Assessment cross-references the business content, business processes, areas of responsibility, managed assets, external clients, related systems, data categories, existing regulations and manuals, educational materials, past legal modification response records, internal audit findings, external audit reports, business contracts, system design documents, operation procedure manuals, customer response manuals, franchise store management ledgers, transaction monitoring rules, risk assessment forms, internal whistleblower records, and past administrative guidance response records of each department, including Service Operations, System Development, Customer Support, Risk Management, Sales, Finance, Legal Affairs, and Audit, with the key points of legal modifications extracted by the generative AI. For example, if the Service Operations Department is responsible for "Personal Confirmation," "Franchise Store Management," "Settlement Service Application," "User Compensation Response," and "New Service Planning," it will be automatically identified as a department affected by strengthened personal confirmation procedures, new user explanation obligations, the introduction of user compensation systems, and legal compliance assessments of new services. For example, if the system development department is responsible for "settlement system development," "API management," "security response," "fault response," "database design," "log management," and "external system integration," then it is considered a department affected by transaction record retention obligations, system specification changes, API specification changes, strengthened security requirements, and fault reporting obligations. For example, if the customer support department is responsible for "customer response," "acceptance of personal confirmation documents," "FAQ response," "complaint handling," "user compensation acceptance," and "personal confirmation re-procedure guidance," then it becomes a department affected by new obligations to explain to customers, changes to the personal confirmation process, and user compensation system responses. For example, if the risk management department is responsible for "AML / CFT response," "transaction monitoring," "legal compliance," "risk assessment," "internal audit," "external audit response," "suspicious transaction reporting," and "risk reduction strategy development," then it becomes a department affected by new AML / CFT requirements, changes to monitoring benchmarks, reassessment of internal audit procedures, and external audit responses. For example, if the sales department is responsible for "franchise development," "service proposals," "customer explanations," "new market research," "franchise review," "contract signing," and "new service briefings," then it needs to address changes to explanatory materials for franchisees and new service proposals, as well as stricter franchise review benchmarks. For example, if the finance department is responsible for "settlement transaction record management," "ledger preservation," "tax declaration," and "responding to the Electronic Ledger Preservation Act," then it becomes a department affected by amendments to the Electronic Ledger Preservation Act and tax system changes. Similarly, if the legal department is responsible for "contract management," "legal investigation," "litigation response," "administrative response," and "internal regulation management," then it needs to revise contract templates and respond to new legal obligations.For example, if the audit department is responsible for "internal audit," "external audit response," "risk assessment," and "audit report preparation," then it needs to add audit projects accompanying legal amendments and reassess audit steps. Specifically, the Department Impact Assessment Department executes a similarity calculation algorithm in a multi-dimensional vector space. This Department Impact Assessment Department obtains the business content text and related document groups of each department from the Database Management Department. The Department Impact Assessment Department inputs the obtained text groups into an embedding model to generate departmental configuration file vectors representing each department. These departmental configuration file vectors are, for example, dense vectors of 248 dimensions. The Department Impact Assessment Department inputs the key text of the legal amendments extracted by the Generative AI Execution Department into the same embedding model to generate key vectors of the legal amendments. The Department Impact Assessment Department calculates the cosine similarity between the key vectors of the legal amendments and the configuration file vectors of each department. The Department Impact Assessment Department normalizes the calculated similarity score and uses it as the impact score for each department. The Department Impact Assessment Department identifies departments whose impact scores exceed a pre-set threshold as affected departments. The input data to this Departmental Impact Assessment Department is a tensor consisting of a vector of key legal amendment points and vectors of multiple departmental configuration files. The output data from this Departmental Impact Assessment Department is an array containing pairs of departmental identifiers and impact scores. This output data includes numerical data such as "System Development Department: Score 0.89" and "Risk Management Department: Score 0.92". The Departmental Impact Assessment Department then sends a list of identified departments to the Response Generation Department. Through this process, the system accurately assesses semantic relationships that simple methods such as keyword matching cannot fully capture. This structure allows for the comprehensive identification of the scope of legal amendment impact across the entire organization. This structure prevents unnecessary notifications to departments without impact, optimizing overall organizational efficiency. In large enterprises with complex matrix organizations, this system clearly defines the boundaries of responsibility between departments, supporting the construction of rapid response mechanisms.

[0088] The response plan generation department uses generative AI to generate prompts for departments identified as affected, such as: "Based on the following key points of legal amendments and the department's business content, create necessary application process change plans, system specification change plans, customer notification plans, existing manual revision plans, new personal confirmation service proposals, franchise store review benchmark revision plans, fault response procedure reassessment plans, user compensation system implementation plans, FAQ update plans, educational material revision plans, internal audit checklist revision plans, external client response plans, API specification change plans, database design change plans, and risk assessment procedure reassessment plans," "Create briefing scripts and FAQ drafts based on the following content," and "Create customer notification templates based on the following content." For example, AI automatically generates prompts for service operations departments such as "reassessment of personal confirmation process," "creation of new user manuals," "revision of franchise store management regulations," "formulation of user compensation system implementation procedures," and "addition of legal compliance assessment process for new services." For system development departments, prompts include "addition of transaction record storage function," "API specification change," "strengthening of security requirements," "installation of automatic fault reporting function," "reassessment of external system linkage specifications," "database design change," and "daily..." The document outlines several key areas of focus, including: strengthening the management of personal confirmation documents; for the customer support department, developing new procedures for accepting personal confirmation documents; updating FAQs; creating customer notification templates; re-evaluating complaint handling procedures; adding user compensation acceptance processes; and creating guidelines for re-confirming personal confirmation procedures. For the risk management department, this includes: implementing new AML / CFT monitoring benchmarks; re-evaluating risk assessment procedures; updating internal audit checklists; revising suspicious transaction reporting benchmarks; developing risk reduction strategies; and re-evaluating external audit response procedures. For the sales department, this includes: providing explanatory materials to franchisees. The tasks included: revision of materials, preparation of new service proposal materials, stricter response to franchisee review standards, re-evaluation of new market research procedures, and preparation of scripts for new service briefings. For the finance department, the tasks included: revision of ledger preservation procedures, re-evaluation of tax filing procedures, and revision of the manual for responding to electronic ledger preservation laws. For the legal department, the tasks included: revision of contract templates, preparation of solutions for new legal obligations, revision of internal regulations, and re-evaluation of administrative response procedures. For the audit department, the tasks included: addition of audit projects, re-evaluation of audit procedures, and revision of audit report formats.For example, regarding new services and business opportunities that become possible due to legal amendments, generative AI also proposes new, feasible business solutions based on the following key points of legal amendments and departmental business content: "Initiate new personal confirmation services," "Develop new markets for electronic settlement transactions," "Expand new services through API linkage," "Provide legal amendment response consulting services," "Develop risk assessment services for franchisees," and "Utilize user compensation systems to acquire new customers." Specifically, the solution generation department executes a conditional text generation algorithm. Based on the department identifier received from the affected department's determination department, the solution generation department retrieves text data from the database management department, including the department's business manual and system design document. The solution generation department combines the retrieved text data with the key points of the legal amendments to generate a context vector. The solution generation department embeds the context vector into the variable portion of the prompt template. The solution generation department inputs the generated prompt into a large-scale language model. The input data to this large-scale language model is a multi-dimensional array of tags containing, for example, "the current API specifications of the system development department" and "new security requirements." This large-scale language model, based on an input array of tags, autoregressively predicts the next tag and generates response text. During the generation process, the model adjusts temperature parameters to balance the diversity and accuracy of the output. The output data from this model is structured JSON text containing specific action items, responsible roles, and deadlines. This output data includes detailed instructions such as "Action: Update the encryption method of the API endpoint," "Object System: Settlement Gateway," and "Priority: High." The response generation department inputs the generated response text into a syntax parsing module to verify the format's appropriateness. The response generation department stores the verified responses in the database management department. Through this process, the system transforms abstract legal amendment content into concrete business tasks that departments can immediately execute. This structure significantly shortens the lead time for the planning phase in legal amendment responses. This structure clearly defines tasks requiring inter-departmental collaboration, improving project management efficiency. This system supports the optimal allocation of development resources in electronic settlement services accompanying various system modifications.

[0089] The system automatically generates legal amendment notices, FAQ drafts, briefing scripts, customer notification templates, educational materials, revised user manuals, revised contract templates, audit checklists, external client response materials, API specifications, database design documents, risk assessment forms, etc., for various departments. Responsible personnel can view, edit, and download these documents through the user interface department and the document viewing and editing department. For example, the service operations department can add company-specific applications to the AI-generated personal confirmation process; the system development department can add company-specific requirements to the API specification change plan; the customer support department can add frequently encountered on-site issues to the FAQ draft; and the sales department can incorporate company advantages into new service proposals. Similarly, the audit department can add company-specific audit items to the AI-generated audit checklist. Specifically, the document viewing and editing department has a rich text editor component and a version management module. Responding to requests from user terminals, the document viewing and editing department retrieves the corresponding document data from the database management department. The document viewing and editing department renders the retrieved data into HTML format and sends it to the user interface department. The user interface department then displays the rendered document on the screen. This document viewing and editing department captures real-time text appending, deletion, and formatting changes performed by users. It uses an operation conversion algorithm to transform these events and synchronizes them with the master data on the server. Whenever an edit is confirmed, the department creates a new version of the document and saves it along with the change differences in the database management department. The department performs a merging process to resolve conflicts when multiple users are simultaneously editing the same document. Through this process, the system provides an environment that seamlessly integrates human expertise with the company's inherent proprietary technology for AI-generated documents as a baseline. This architecture eliminates the need for a zero-to-zero initiation process for document creation, allowing human resources to be focused on review and refinement. This architecture enables complete tracking of change history, ensuring an audit trail of who modified which part and when. In compliance departments of financial institutions with stringent document management requirements, this system achieves both improved business quality and reduced processing time.

[0090] The company also utilizes AI-generated materials for briefings and internal training sessions, as well as briefings for franchisees, external clients, and the group company. For example, the service operations department holds an internal briefing on the new personal confirmation process, the customer support department conducts training on new customer response procedures, the sales department holds a new service briefing for franchisees, the finance department conducts an internal training on the amendments to the Electronic Ledger Preservation Act, the legal department holds a company-wide briefing on new legal obligations, and the audit department holds a briefing on changes to the audit process. Following the briefing and in daily operations, when a user inputs a question from their terminal, the chatbot response department, powered by generative AI on the server, provides immediate answers to questions such as "What are the specific details of the new personal confirmation requirements?", "Under what circumstances does the obligation to retain transaction records apply?", "When should customer notifications be made?", "What are the specific details of the API specification changes?", "What is the basis for adding audit projects?", "What is the scope of the user compensation system?", "How should we respond to external clients?", and "What are the legal considerations for expanding into new markets?". Specifically, this chatbot response department is installed as a dialogue system with session management capabilities and a memory network. When it receives a question text from a user's terminal, it identifies the user's session ID. Based on the session ID, it retrieves past conversation history and the user's departmental profile from the database management department. Combining the question text, conversation history, and departmental profile, it generates context-aware prompts. These prompts are then input into a large-scale language model. The input data to this large-scale language model is a text tensor containing questions such as "What are the specific details of the API specification change?" from users in the system development department. Based on the input tensor, this large-scale language model generates response text specialized to the company's internal system architecture. The output data from this model is structured data containing the response text and hyperlinks to relevant internal company documents. This output data includes specific instructions such as "The validity period of the authentication token needs to be shortened. Please refer to the design document of the linked target for details." The chatbot response department sends the output response data to the user's terminal. Through this processing, the system provides highly personalized responses based on an understanding of the user's attributes and past context. This architecture immediately solves complex practical questions that general FAQ systems cannot handle. This architecture significantly reduces the number of upgrades required from the legal department, preventing company-wide business disruptions. This system functions as a hub for information sharing in large-scale legal amendment projects involving multiple stakeholders.

[0091] In addition, the Log Management Department records the system's full processing history, user operation history, and AI output history. For example, it records in detail who uploaded which legal amendment document and when, which department viewed and edited which response plan, what AI answer was given to which question, who received which notification, how which data was corrected, when which progress was completed, which external client responded to which issue, and which group company referenced which data. The Notification Management Department automatically distributes legal amendment information and response plans, data updates, briefing guides, FAQ updates, and reminders for unresolved tasks to various terminals. For example, it can instantly receive notifications such as "New legal amendment information added," "This department has been selected as an affected department," "New response plan generated," "Briefing guide for holding a briefing," "FAQ updated," "Unresolved tasks," "Response to external clients incomplete," and "Data distribution to group companies completed." The History Management Department records the response content and progress of past legal amendments in chronological order, which can be used as reference information for future legal amendments. For example, the log management system records, in chronological order, which departments responded to past legal amendments, which documents were revised, what issues arose, what new business proposals were made, which external clients responded to which issues, and which group companies referenced which documents. Specifically, to ensure observability in the distributed system environment, the log management department employs structured logging and distributed tracing technologies. This department collects event data output from each microservice as JSON-formatted log entries. These log entries include timestamps, user IDs, session IDs, executed API endpoints, and processing status codes. The collected log entries are forwarded to the log aggregation infrastructure, indexed, and stored in a searchable state. The notification management department employs an asynchronous communication architecture using a message broker. It uses specific events occurring within the system as triggers to publish notification messages to a message queue. Subscriber processes retrieve messages from the queue and select the optimal distribution channel based on user notification settings. The historical record management department builds a data warehouse to store accumulated log data and historical response data in a multi-dimensional analytical format. This historical record management department applies machine learning algorithms to past project data to build models that predict task delay risks and signs of resource shortages. Through this process, the system fully visualizes the overall operational status, enabling rapid troubleshooting. This structure pushes and distributes information to the appropriate users at the right time, preventing task postponement and oversights. This structure does not treat past data as mere records, but rather utilizes it as predictive data to optimize future project management. This system strongly supports continuous process improvement within the organization while meeting stringent auditing requirements.

[0092] Furthermore, even when multiple legal amendments occur simultaneously, generative AI independently analyzes the content of each amendment, automatically adjusting for overlaps in affected departments and integrating response plans. For example, if the Payment Services Act and the Personal Information Protection Act are amended simultaneously, it can suggest combined application changes and data correction plans for departments affected by both amendments. Additionally, even when departments have different business processes and management systems, generative AI learns each department's business processes and past application history to individually generate optimal response plans. For example, even with the same legal amendment, it can individually suggest revisions to the customer confirmation process for the service operations department, API specification changes for the system development department, FAQ updates for the customer support department, new monitoring benchmark implementation for the risk management department, new service proposal material creation for the sales department, revised ledger maintenance procedures for the finance department, revised contract templates for the legal department, and additional audit projects for the audit department. Specifically, to comprehensively process multiple legal amendment information, this generative AI execution unit uses graph neural network dependency modeling. It constructs a knowledge graph with key texts of each legal amendment as nodes and semantic relationships between them as edges. The unit updates node features using graph convolutional networks, learning the interactions between legal amendments. The response generation unit receives the updated feature vectors as input and solves the constraint satisfaction problem of simultaneously meeting multiple legal regulatory requirements. The input data to the response generation unit is a multidimensional tensor combining, for example, the "funds settlement law amendment vector" and the "personal information protection law amendment vector." Considering both vectors, the response generation unit generates a non-contradictory composite response text. The output data from this model is structured data containing a single business process change plan that integrates responses to multiple legal regulations. This output data includes specific actions such as "integrating the consent process for obtaining personal information upon personal confirmation." The response generation unit also maintains a state transition model representing the inherent business processes of each department. This system simulates and determines the feasibility of response plans generated by the response plan generation department within the state transition models of each department. Through this process, the system seamlessly integrates multiple complex and intertwined legal regulatory requirements, preventing on-site confusion. Based on a deep understanding of the context of each department, this structure allocates the most efficient and achievable tasks. This structure fully automates the manual verification of legal regulations performed by the legal department. In today's increasingly complex regulatory business environment, this system dramatically improves a company's compliance capabilities.

[0093] Additionally, the system can also open some functions to external clients and group companies to enhance the coordination of responses to legal amendments. External clients can input their response status from the terminal or refer to AI-generated materials. For example, it can automatically generate "new contract templates for external clients," "application change plans for group companies," "educational materials for external clients," and "FAQs for group companies," and share them among relevant personnel. Specifically, this system implements a multi-tenant architecture and a role-based access control mechanism. The system constructs a tenant environment that is logically separated by external clients and group companies. The database management department applies row-level security to ensure data isolation so that each tenant can only access its own data. The user interface department dynamically switches the displayed menus and operable functions based on the logged-in user's role information. When generating materials for external clients, the response solution generation department assigns the context parameter "for external clients" to the prompt. The input data input to the response solution generation department is a tensor containing a vector of key points of legal amendments and a vector of contractual conditions with external clients. The output data from the response generation department is document text with confidential information masked and externally tailored for presentation. This output data includes text such as "compliance steps for new security benchmarks in outsourced business." The system distributes the generated documents to the external client's terminals via a secure file-sharing protocol. The system receives response completion reports from external clients via API and automatically updates the progress status of the database management department. Through this process, the system is not limited to a single enterprise but builds a comprehensive compliance management system involving the entire supply chain. This structure reduces communication costs with external clients and prevents delays and discrepancies in information delivery. This structure strengthens overall group governance and effectively reduces third-party risks. This system has demonstrated extremely high adoption rates among platform operators that collaborate with numerous partner companies to provide services.

[0094] Thus, in this embodiment of the invention, generative AI must be effectively utilized in the main processing unit on the server side to realize user experience such as user interface and chatbot response, notification reception, data editing, and progress management on the terminal side. This unifies and efficiently automates the entire process of responding to legal amendments in the electronic settlement service field, reducing the risk of omissions and legal violations while promoting the discovery and proposal of new business opportunities arising from legal amendments, achieving a balance between defensive and offensive responses. Specifically, this system integrates advanced natural language processing models and distributed computing technology to fundamentally reconstruct legal compliance operations that previously relied on advanced human cognitive abilities using computer science methods. This system pipelines and executes a series of algorithms, including structuring unstructured data, calculating semantic similarity in high-dimensional vector spaces, and generating conditional text. By parallelizing these processes, the system reduces the latency of parsing large amounts of documents to milliseconds. By directly inputting the inference results of machine learning models into the subsequent business workflow engine, this system achieves end-to-end automation without human intervention. This system is not merely a tool for improving business efficiency; rather, it's an information processing device that captures changes in the external environment—such as alterations to legal regulations—as updates to the enterprise's internal data structure, and autonomously optimizes the overall system's state. Through this structure, the system reduces the human resources and time required to respond to legal amendments to a fraction of what was previously necessary. This structure eliminates human oversights and interpretive biases, mathematically guaranteeing the accuracy of compliance responses. By using legal constraints as parameters for new business operations, this structure enables data-driven business development. In a complex legal regulatory environment, this system provides an indispensable technological foundation for enterprises to achieve sustainable growth.

[0095] (System Composition) The system described in this embodiment includes a legal amendment information acquisition unit, an affected department determination unit, a response plan generation unit, a database management unit, a generative AI execution unit, a log management unit, a notification management unit, a history management unit, a user interface unit, a data viewing and editing unit, a chatbot response unit, a notification receiving unit, and an operation log sending unit. Specifically, this system is configured as a distributed processing system based on a cloud-native architecture. Each processing unit constituting this system is installed as a containerized microservice and runs on a container orchestration platform. The server is a collection of physical or virtual computing nodes equipped with multi-core processors, large-capacity volatile memory, and high-speed solid-state drives. To accelerate the inference processing of large-scale language models, the generative AI execution unit is configured on dedicated computing nodes equipped with multiple image processing semiconductor devices specialized for tensor operations. The database management unit employs a hybrid structure combining relational and NoSQL databases, providing the optimal storage hierarchy corresponding to the characteristics of the data. The terminal is any information processing device capable of running a web browser and is connected to the server via a secure encrypted communication channel. This architecture allows the system to flexibly scale computing resources horizontally in response to changes in processing load. It eliminates single points of failure, achieving high availability and fault tolerance. The system provides a robust infrastructure that simultaneously meets enterprise-level security and performance requirements.

[0096] The Legal Amendment Information Acquisition Department is capable of receiving various legal amendment-related documents uploaded from terminals by legal personnel, risk management personnel, system developers, customer support personnel, and sales personnel. These documents include amendments to laws such as the Funds Settlement Act, the Act on Prevention of Transfer of Proceeds of Crime, Consumer Protection Act, the Electronic Bookkeeping Act, tax system amendments, and the Personal Information Protection Act. The department also receives these documents along with guidelines, notices from the Executive Yuan, industry group opinions, precedents, administrative guidance documents, public opinions, and related news reports. Furthermore, the department automatically retrieves information from external legal databases or government agency websites via API linkage or web scraping. For example, it can collect, in a unified manner, PDF format amendments to the Funds Settlement Act, HTML format guidelines, Word documents containing industry group opinions, administrative notices automatically retrieved from the Executive Yuan website, summaries of public opinions, newspaper clippings, explanatory reports from industry magazines, Executive Yuan FAQs, and past administrative guidance examples. The Legal Amendment Information Acquisition Department transmits received or acquired documents to the Generative AI Execution Department, generating prompts such as "Summarize the following amendments to the Payment Services Act and extract the main amendments, implementation dates, additional obligations, strengthening of penalties, and transitional measures," "Summarize the Executive Yuan's interpretation policy, practical considerations, and impact on electronic settlement services from the following full text of guidelines," "Summarize the main opinions and the Executive Yuan's responses from the following public opinion results," and "Extract the envisioned questions and answers from the following Executive Yuan FAQs," and instructs the Generative AI to analyze them. Specifically, this Legal Amendment Information Acquisition Department functions as an advanced data pipeline module performing extraction, transformation, and loading processes. In communication with external systems, this Legal Amendment Information Acquisition Department establishes secure data linkage using RESTful APIs and OAuth authentication protocols. For the acquired unstructured data, this Legal Amendment Information Acquisition Department performs data cleaning processes such as character encoding standardization, removal of unnecessary HTML tags, and noise removal using regular expressions. The Legal Amendment Information Acquisition Department uses a natural language processing library to segment the cleaned text data into blocks, either sentence-by-sentence or paragraph-by-paragraph. This department assigns metadata (source URL, acquisition date and time, document category, etc.) to each segmented block, generating a structured JSON object. The department then combines this generated JSON object with a prompt template containing extraction instructions. This combined prompt string is sent to the inference API of the Generative AI Execution Department. The data input to the Generative AI Execution Department is a string containing, for example, "text blocks containing amendments to the Funds Settlement Agreement." The output data from the Generative AI Execution Department is structured data composed of extracted key items as key-value pairs. This output data includes data such as "Effective Date: April 1, 2024" and "Additional Obligation: Preservation of Transaction Records."This process transforms diverse external data with inconsistent formats into a standardized data format that can be uniformly processed within the system. This structure reduces noise in subsequent AI parsing, significantly improving extraction accuracy. It also fully automates data collection, greatly reducing the workload for legal personnel. In compliance operations where timeliness and accuracy of information are paramount, this system provides a robust data foundation.

[0097] The Department Impact Assessment Unit has the capability to cross-reference the business content, business processes, areas of responsibility, managed assets, external clients, related systems, data categories, existing regulations and manuals, educational materials, past legal modification response records, internal audit findings, external audit reports, business contracts, system design documents, operation procedures, customer response manuals, franchise store management ledgers, transaction monitoring rules, risk assessment forms, internal whistleblower records, and past administrative guidance response records of various departments, including Service Operations, System Development, Customer Support, Risk Management, Sales, Finance, Legal Affairs, and Audit, with the key points of legal modifications extracted by generative AI. It analyzes the correlation between each department's business content and the legal modification content from multiple perspectives, automatically identifying the affected departments and prioritizing those with high impact. For example, if the Service Operations Department is responsible for "Personal Confirmation," "Franchise Store Management," "Settlement Service Application," "User Compensation Response," and "New Service Planning," it will be automatically identified as a department affected by strengthened personal confirmation procedures, new user explanation obligations, the introduction of user compensation systems, and legal compliance assessments of new services. For example, if the system development department is responsible for "settlement system development," "API management," "security response," "fault response," "database design," "log management," and "external system integration," then it is identified as a department affected by transaction record retention obligations, system specification changes, API specification changes, enhanced security requirements, and fault reporting obligations. Similarly, if the customer support department is responsible for "customer response," "acceptance of personal confirmation documents," "FAQ response," "complaint handling," "user compensation acceptance," and "personal confirmation re-procedure guidance," then it is identified as a department affected by new obligations to explain to customers, changes to the personal confirmation process, and user compensation system responses. Furthermore, if the risk management department is responsible for "AML / CFT response," "transaction monitoring," "legal compliance," "risk assessment," "internal audit," "external audit response," "suspicious transaction reporting," and "risk reduction strategy development," then it is identified as a department affected by new AML / CFT requirements, changes to monitoring benchmarks, reassessment of internal audit procedures, and external audit responses. As a suggestion for generative AI, one could be: "Check the following key points of legal amendments against the business content databases of each department, list the affected departments, and extract highly relevant business processes and managed assets." Specifically, this Departmental Impact Assessment Unit executes clustering and classification algorithms in a multi-dimensional vector space. It obtains text data representing the business content of each department from the Database Management Unit and converts it into departmental configuration file vectors using a pre-learned language model. Similarly, this Departmental Impact Assessment Unit converts the text of legal amendment key points received from the Generative AI Execution Unit into legal amendment key point vectors.This Department Impact Assessment Unit uses machine learning classifiers such as Support Vector Machines or Random Forests to probabilistically determine which department's business domain the key points of the legal amendment belong to. The input data to this department consists of a feature matrix composed of the key points of the legal amendment and the configuration file vectors of each department. The output data from this department is an array of impact probability scores for each department. This output data includes probability values ​​such as "System Development Department: 0.95" and "Customer Support Department: 0.82". The Department Impact Assessment Unit sorts the calculated probability scores in descending order and lists departments exceeding a predetermined threshold as impact departments. This department then sends the list of identified impact departments and identifiers of highly relevant business processes to the Response Generation Unit. Through this process, the system eliminates subjective human judgment, achieving objective and quantitative impact assessment based on data. This structure accurately identifies the scope of the impact of the legal amendment, preventing omissions in the response. This structure makes it possible to prioritize resource allocation to departments with high impact. In large organizations with highly segmented business operations, this system accelerates the initial initiation of cross-departmental projects.

[0098] The response plan generation department uses generative AI to generate prompts for departments identified as affected, such as: "Based on the following key points of legal amendments and departmental business content, create necessary application process change plans, system specification change plans, customer notification plans, existing manual revision plans, new self-confirmation service proposals, franchise store review benchmark revision plans, fault response procedure reassessment plans, user compensation system implementation plans, FAQ update plans, educational material revision plans, internal audit checklist revision plans, external client response plans, API specification change plans, database design change plans, and risk assessment procedure reassessment plans," "Create briefing scripts and FAQ drafts based on the following content," and "Create customer notification templates based on the following content." For example, for the service operations department, AI automatically generates prompts such as "Reassessment of self-confirmation process," "Creation of new user manuals," "Revision of franchise store management regulations," "Formulation of user compensation system implementation procedures," and "Addition of legal compliance assessment process for new services." For the system development department, it automatically generates prompts such as "Addition of transaction record saving function," "API specification change," "Enhancement of security requirements," "Installation of automatic fault reporting function," "Reassessment of external system linkage specifications," "Database design change," and "Enhancement of log management function." The system automatically generates tasks such as "formulating new procedures for accepting personal confirmation documents," "updating FAQs," "creating customer notification templates," "re-evaluating complaint handling procedures," "adding user compensation acceptance processes," and "creating guides for personal confirmation re-procedures" for the customer support department. It also automatically generates tasks such as "importing new AML / CFT monitoring benchmarks," "re-evaluating risk assessment procedures," "updating internal audit checklists," "revising suspicious transaction reporting benchmarks," "formulating risk reduction strategies," and "re-evaluating external audit response procedures" for the risk management department. Finally, it automatically generates tasks such as "revising explanatory materials for franchisees" for the sales department. The system automatically generates various tasks for the finance department, including tasks such as "ordering," "creating new service proposal materials," "responding to stricter franchisee review standards," "re-evaluating new market research procedures," and "creating scripts for new service briefings." For the finance department, it automatically generates tasks such as "revising ledger preservation procedures," "re-evaluating tax filing processes," and "revising the electronic ledger preservation law response manual." For the legal department, it automatically generates tasks such as "revising contract templates," "creating solutions for new legal obligations," "revising internal regulations," and "re-evaluating administrative response procedures." For the audit department, it automatically generates tasks such as "adding audit projects," "re-evaluating audit procedures," and "revising audit report formats." Specifically, this response generation department executes a text generation pipeline with a feedback loop utilizing reinforcement learning. The response generation department incorporates departmental identifiers and key legal amendment data received from the influencing department into prompt templates. The generated prompts are then input into a large-scale language model.The input data to this large-scale language model consists of a vector representing the current business process state of the department and a tensor representing the constraints imposed by legal changes. Based on the input tensor, this large-scale language model generates a sequence of text tags representing the steps involved in the business process changes. The output data from this model is structured JSON-formatted text containing task names, roles, and specific operational steps. This output data includes data such as "Task: Update API Specification" and "Step: Change the endpoint authentication method to OAuth2.0". To evaluate the quality of the generated text, the response generation department has a module that automatically calculates evaluation metrics such as BLEU scores or ROUGE scores. The response generation department uses past user modification history data as a reward signal to fine-tune the weight parameters of the generation model. Through this process, the system becomes increasingly capable of generating highly accurate responses tailored to the company's business processes with repeated use. This structure specializes a general AI model for a specific enterprise domain, maximizing its practical value. This structure dramatically reduces the time required to create a response from several days to several minutes. This system accelerates the initial response to legal amendments, becoming a powerful engine for enterprises to respond swiftly to regulatory changes.

[0099] The Database Management Department centrally manages all data related to legal amendments, including departmental business content, past response records, generated summaries, response plans, materials, FAQs, scripts, Q&A, educational materials, contract templates, audit checklists, external client response materials, API specifications, database design documents, risk assessment forms, etc. For example, it accumulates past legal amendment-related materials such as application manual revision plans, briefing materials, FAQs, Q&A records, external client response records, and group company response records for future reference during legal amendments. Specifically, the Database Management Department employs a hybrid storage architecture for comprehensively handling both structured and unstructured data. For transactional data such as user information and permission settings, the department uses a relational database that guarantees ACID properties. For large-volume, schema-less data such as generated documents and log data, the department uses a distributed NoSQL database. To achieve high-speed text retrieval, the department integrates a full-text search engine with an inverted index structure. The department also has a vector database module that stores vectorized document data, enabling near-nearest neighbor searches. This database management department dynamically updates indexes using B-trees or log-structured merged trees during data writes. To prevent data loss, the department performs backup processing through regular snapshots and asynchronous replication. Through this process, the system securely and efficiently manages petabytes of massive legal and business data. This structure guarantees millisecond-level response times to user retrieval requests. This structure preserves the company's past response history—its knowledge assets—by persisting it as a knowledge base for future decision-making. This system functions as a robust data foundation for achieving data-driven compliance operations.

[0100] The generative AI execution department, based on prompts and data received from the legal amendment information acquisition department, the affected department determination department, and the response solution generation department, performs natural language processing and knowledge reasoning to generate various outputs such as summaries, extractions, analyses, proposals, document generation, FAQ generation, script generation, and Q&A. For example, it performs various generation tasks such as "extracting key points of legal amendment provisions," "summarizing practical explanations of guidelines," "analyzing the correlation between business content and legal amendment content," "automatically generating change application solutions," "proposing new business opportunities," "automatically creating briefing scripts," "automatically generating FAQs," and "chatbot-based Q&A." Input prompts to the generative AI include: "Summarize the following legal amendment-related documents and extract the main amendment points, implementation date, additional obligations, penalties, and transitional measures," "Check the following departmental business content against the key points of the legal amendments and list the affected departments and necessary response solutions," "Create a briefing script and FAQ based on the following content," and "Answer the following questions based on the key points of the legal amendments and departmental business content." Specifically, this generative AI execution department contains a large-scale language model with a self-attention mechanism. This model is constructed as a deep neural network with hundreds of billions of parameters. The processing unit uses a tokenizer to convert the input text data into a fixed-length sequence of tokens. This processing unit maps each token to a high-dimensional embedding vector. The data input to this model is a text vector containing prompts and related documents. The dimension of this input data is, for example, 4096 dimensions. The elements of this input data contain distributed representations of words and positional encoding information. As part of this preprocessing, this processing unit performs the removal of unwanted symbols and text normalization. Specific examples of this input data include: first, text data of legal amendments; second, text data of business manuals from various departments; and third, natural language-based question text from users. This model applies a multi-layered Transformer encoder and decoder to the input vector sequence. This model calculates contextual dependencies. The output data from this model is a probability distribution vector of the words to be generated next. The format of this output data is an array of floating-point numbers with a dimension equal to the size of the vocabulary. Each output value represents the probability of a specific word appearing next. As specific examples of the output data, firstly, the key text of the summarized legal amendments can be provided. Secondly, a list of affected departments can be provided. Thirdly, the text data of the generated FAQ can be provided. This follow-up processing unit applies a cluster search algorithm based on the probability distribution of the output. This follow-up processing unit determines the text sequence with the highest probability. This follow-up processing unit forwards the generated text to the response generation unit or the user interface unit. Through this structure, this system replaces manual document parsing performed by human experts. This system significantly improves the speed of document processing. This system effectively improves the accuracy of information processing in the initial stages of responding to legal amendments.This technology can be applied to use cases in the legal technology field that rapidly parse large legal documents and automatically generate practical solutions.

[0101] The Log Management Department records the entire processing history of the system, user operation history, and AI output history. For example, it records in detail who uploaded which legal amendment document and when, which department viewed and edited which response plan, what AI answer was given to which question, who received which notification, how which document was corrected, when which progress was completed, which external client responded, and which group company referenced which document. Specifically, this Log Management Department uses a distributed database system. This processing department collects all events occurring within the system via asynchronous message queues. This processing department assigns a unique timestamp to the collected event data. This processing department formats the event data into structured log records. This processing department applies cryptographic hash functions to the log records. This processing department includes the hash value of the immediately preceding log record in the current log record. This processing department constructs a hash chain. Through this structure, the system makes log data tampering mathematically detectable. This processing department stores log data in a time-series database. This processing department indexes the log data. This processing department enables high-speed retrieval and querying. This processing department periodically compresses the log data. This processing unit archives compressed log data in inexpensive cold storage. This architecture optimizes storage capacity consumption. This system enables the maintenance of long-term audit trails. This system leverages the efficiency of data management and enhances security. This technology can be applied to audit response systems in industries with stringent compliance requirements, such as financial institutions or healthcare organizations.

[0102] The Notification Management Department automatically distributes information on legal amendments and corresponding solutions, updated materials, briefing session guides, FAQ updates, and reminders for unresolved tasks to all terminals. For example, it can instantly receive notifications such as "New legal amendment information added," "This department has been selected as an affected department," "New response solutions generated," "Briefing session guide," "FAQ updated," "Unresolved tasks," "Incomplete responses to external clients," and "Materials distributed to the group company completed." Specifically, the Notification Management Department adopts a publish-subscribe messaging architecture. The Processing Department receives status change events from various modules within the system. The Processing Department parses the received events. The Processing Department determines the appropriate notification subject based on the event type. The Processing Department refers to each user's attributes and department information. The Processing Department dynamically identifies the user group as the destination of the notification. The Processing Department generates notification messages for the identified user group. The Processing Department utilizes the WebSocket communication protocol. The Processing Department establishes a continuous two-way communication channel with the client terminals. The Processing Department pushes and distributes the generated notification messages to the client terminals in real time. The Processing Department detects when a client terminal is offline. This processing unit maintains unsent notification messages in a temporary queue. The processing unit resends notification messages when the client terminal returns to online status. Through this architecture, the system implements a robust notification mechanism that is highly resistant to network delays or disconnections. This system delivers important legal changes and task updates to users without delay. This system effectively prevents business delays and improves the overall response speed of the organization. This technology can be applied to use cases in large organizations for building internal communication infrastructures that reliably and rapidly communicate highly urgent information to a large number of employees.

[0103] The Historical Records Management Department records the responses and progress of past legal amendments in a chronological order, which can be used as reference information for future legal amendments. For example, it records which department responded to past legal amendments, what documents were revised, what questions were raised, what new business proposals were made, which external clients handled which responses, and which group companies referenced which documents. Specifically, this Historical Records Management Department has a multi-dimensional time-series database engine. This processing department stratifies and stores data according to each past response item. This processing department assigns metadata to each response action. This metadata includes the executor ID, execution date and time, object law ID, and impact score. This processing department applies a natural language processing model to the stored historical record data. This processing department extracts feature vectors from the historical record data. This processing department stores the extracted feature vectors in a vector database. This processing department detects when new legal amendment information is input. This processing department similarly generates a query vector from the new legal amendment information. This processing department calculates the cosine similarity between the query vector and the feature vectors in the vector database. This processing department extracts past response records with similarity exceeding a predetermined threshold. This processing department provides the extracted past response history as reference information to the user interface department. Through this structure, the system makes past tacit knowledge reusable as quantitative data. This system significantly reduces the time required to develop response plans for similar legal changes. By automatically highlighting past successes and failures, the system improves the quality of responses. This technology can be applied to knowledge management systems within corporate legal and compliance departments, accumulating and utilizing past response know-how as an organizational knowledge base.

[0104] The User Interface (UI) department provides access to the server from terminals for responsible personnel in various departments to view and download generated summaries, lists of affected departments, response plans, notification materials, FAQs, briefing scripts, customer notification templates, educational materials, contract templates, audit checklists, external client response materials, API specifications, database design documents, risk assessment forms, etc. For example, it also provides data retrieval and filtering functions, customized display by department, feedback input and submission of correction plans for AI-generated data, and dedicated screens for external clients and the group company. Specifically, this UI department is built as a single-page application using the Virtual Document Object Model (VDOM). This processing department performs asynchronous RESTful API communication with the server. This processing department dynamically updates content without reloading the entire screen. This processing department evaluates access control lists based on user authentication information. This processing department renders dedicated views corresponding to the user's permission level. This processing department applies virtual scrolling technology to efficiently display large amounts of data. This processing department only expands data elements existing in the visible area of ​​the screen into memory. This processing department receives search queries from users. This processing unit sends queries to a full-text search engine using an inverted index. It retrieves the search results in milliseconds and displays them on the screen. This processing unit collects user feedback on AI-generated data. It structures the feedback data and sends it to the server. This structure reduces the memory consumption of client terminals. This system achieves high-speed rendering and responsiveness for large amounts of data. This system effectively reduces the user's operational burden and improves business efficiency. This technology can be applied to enterprise portals requiring complex access control and large-scale document management.

[0105] The document viewing and editing department provides editing functions that allow users to utilize generated materials as is or to supplement and modify them by department. For example, the service operations department can add descriptions of the company's unique applications to the AI-generated personal confirmation process plan; the system development department can add the company's inherent system requirements to the API specification change plan; the customer support department can add frequently asked questions in the FAQ draft; and the sales department can add the company's advantages to the new service proposal materials. The audit department can also add the company's unique audit items to the AI-generated audit checklist. Specifically, this document viewing and editing department implements a real-time collaborative editing protocol. This department allows multiple users to edit the same document simultaneously. This department uses an operation conversion algorithm to unify the editing operations of each user. This department automatically resolves editing conflicts. This department stores the document's change history as differential data in the version control database. This department applies the Myers differential algorithm. This department calculates the minimum edit distance between the text before and after the change. This department accurately identifies the added and deleted text. This department visually highlights the identified differential information. This processing unit distinguishes between the original text generated by the AI ​​and the text corrected by the user. It extracts the user's corrections as learning data. The extracted data is then appended to the micro-call dataset of the generative AI. Through this structure, the system ensures data consistency in collaborative work performed by multiple people. The system constructs a loop that feeds back the user's tacit knowledge to the system. This system continuously improves the accuracy of future AI-generated outputs. This technology can be applied to use cases such as document management systems for legal documents and technical specifications, where experts from multiple departments jointly review and correct them.

[0106] After briefings or during daily operations, when a user enters a question from the terminal, the chatbot response department sends the content to the server. The generative AI on the server, referencing the question content, key points of legal amendments, departmental business content, and past Q&A history, generates the best answer and returns it to the terminal. For example, it provides immediate answers to questions such as "What are the specific details of the new personal confirmation requirements?", "Under what circumstances does the obligation to retain transaction records apply?", "When should notifications to customers be made?", "What are the specific details of the API specification changes?", "What is the basis for adding audit projects?", "What is the scope of the user compensation system?", "How should we respond to external clients?", and "What are the legal considerations for expanding into new markets?". As a sample prompt to the generative AI, it could be: "For the following questions, based on key points of legal amendments and departmental business content, provide a practical and specific answer." Specifically, this chatbot response department employs a retrieval-enhanced generative architecture. The processing department receives the question text from the user. The processing department inputs the question text into a natural language processing model. The data input to this model is a numerical vector representing the tokenized question text. For example, the input data has 768 dimensions. The elements of this input data are floating-point numbers representing the intent of the question and its context. As part of this preprocessing, this processing unit performs stop word removal and stemming. Specific examples of this input data include: first, a question vector regarding personal confirmation requirements; second, a question vector regarding transaction record keeping obligations. This processing unit uses the question vectors to retrieve data from a vector database. This processing unit extracts relevant legal amendments and past Q&A history based on cosine similarity. This processing unit combines the extracted contextual information with the question text. This processing unit inputs the combined data into a generative AI model. The output data from this model is a probability distribution of the tags constituting the answer text. This output data is in the form of a multidimensional tensor. Each output value represents the probability of generating each word in the answer. Specific examples of this output data include: first, text data explaining the specific personal confirmation steps; second, text data representing the basis for notification timing. This post-processing unit formats the generated answer text. This post-processing unit sends the answer text to the client terminal. Through this architecture, the system achieves advanced information retrieval that understands context, going beyond simple keyword searches. This system effectively addresses user inquiries instantly and prevents business disruptions. This technology can be applied to use cases of conversational AI systems that automate internal help desk and customer support operations.

[0107] The notification receiving department receives notifications from the server in real time and notifies the responsible parties via pop-ups, emails, and chat messages. For example, it can instantly receive notifications such as "New legal amendment information added," "This department has been selected as an affected department," "New response plan generated," "Meeting preparation guide," "FAQ updated," "Unresolved tasks exist," "Response to external clients incomplete," and "Document distribution to the group company completed." Specifically, this notification receiving department resides as a background process on the client terminal. This processing department monitors the persistent connection established with the server. This processing department intercepts data packets pushed and distributed from the server. This processing department decodes the payload of the received data packets. This processing department evaluates the importance flag of the notification contained in the payload. This processing department dynamically changes the display format of the notification based on the importance flag. This processing department calls the operating system's native notification API. This processing department draws a pop-up window at the top of the screen. This processing department registers a callback function when the user clicks the notification. This processing department uses the callback function to directly jump the user to the corresponding document or task screen. This processing department detects when the network connection is temporarily disconnected. This processing department attempts to reconnect using an exponential backoff algorithm. This system ensures notification delivery even in unstable communication environments. It appropriately draws the user's attention. This system effectively prevents the omission of important tasks and supports the efficient execution of business operations. This technology can be applied to push notification systems in mobile or desktop environments to reliably deliver high-urgency alerts to users.

[0108] The operation log sending unit sends the operation history from the terminal to the server, unifying the management system's utilization status and response progress. For example, it records in detail which responsible party viewed and edited which document, received which notification, entered which question into the AI, approved or rejected which response plan, which external client handled which response, and which group company referenced which document. It can automatically calculate response progress based on the operation history and identify departments and tasks that haven't been addressed. Specifically, this operation log sending unit is linked to all event listeners in the user interface on the client terminal. This processing unit captures minute operation events such as mouse clicks, keystrokes, scrolling, and screen dwell time. This processing unit temporarily stores the captured event data in a local memory buffer. This processing unit detects when the amount of data in the buffer reaches a predetermined threshold or when a predetermined time has elapsed. This processing unit serializes the stored data into a JSON-formatted payload. This processing unit applies a data compression algorithm to the payload. This processing unit sends the compressed data to the server as an asynchronous HTTP request. After confirming successful transmission, this processing unit clears the local buffer. Through this configuration, the system minimizes network bandwidth consumption. While reducing the processing load on client terminals, the system collects comprehensive operation logs. Based on the collected log data, the system calculates the task completion rate of each department in real time. The system automatically identifies bottleneck departments and alerts managers. This technology can be applied to use cases such as large-scale system utilization analysis or task management systems for quantitatively evaluating employee performance.

[0109] In this way, the system involved in this implementation, through the coordinated operation of various departments, automates the entire process of responding to legal amendments in the field of electronic settlement services in a unified and efficient manner. Furthermore, by effectively utilizing generative AI, it can comprehensively and specifically support all business operations, including the collection and analysis of legal amendment information, the identification of affected departments, the generation of response plans, document preparation, briefing operation, Q&A, progress management, notification distribution, historical record management, and external collaboration. Specifically, this system runs in a distributed computing environment based on a microservice architecture. Each functional module is deployed as an independent container. The system uses container orchestration tools to dynamically optimize the resource allocation of each module. The system manages a series of data flows from the input of legal amendment information to the final completion of the response as an event-driven state machine. The system executes data transfer between steps at high speed via an in-memory data grid. The system offloads the inference processing of generative AI to a parallel computing cluster using GPUs. Through this configuration, the system performs natural language processing tasks requiring massive computing resources with low latency. The system uses computer science algorithms to systematize and automate the complex legal interpretation and business impact assessment processes that rely on human cognitive abilities. This system effectively eliminates human error, dramatically increases processing speed, and enhances the entire organization's compliance capabilities. This technology can be applied to comprehensive risk management platforms in highly regulated industries such as finance, healthcare, and telecommunications, which face frequent changes in legal regulations.

[0110] (Implementation steps) Step 1: Collection and input of information related to legal amendments Legal, risk management, system development, customer support, and sales personnel upload amendments, guidelines, notices from the Executive Yuan, industry group opinions, precedents, administrative guidance documents, public opinions, and related news materials from terminals to the Funds Settlement Act, the Act on Prevention of Transfer of Proceeds of Crime, Consumer Protection Act, Electronic Ledger Preservation Act, Tax System Amendments, and the Personal Information Protection Act. Additionally, data can be automatically retrieved from external legal databases or government agency websites via API linkage or web scraping. For example, various document formats are available, including PDF amendments, HTML guidelines, Word documents of industry group opinions, administrative notices automatically retrieved from the Executive Yuan website, summaries of public opinions, and clippings from news reports. Specifically, this system initiates a multimodal data import pipeline. The processing unit receives binary data streams from user terminals. The processing unit parses file extensions and magic numbers. The processing unit automatically determines the data format. The processing unit applies an optical character recognition engine to PDF files. The processing unit converts the image-based text into machine-readable string data. This processing unit parses the Document Object Model (DOM) tree of HTML files. It filters out unnecessary decorative tags and navigation elements, extracting only the main text. During external website crawling, it controls a headless browser. It captures text after rendering dynamically generated web pages. It performs character encoding standardization and normalization on all extracted text data. Through this structure, the system converts unstructured data with different formats into a unified structured text format. This system builds the foundation for subsequent natural language processing models that can accurately interpret the data. This system leverages the technology of automating data input to reduce processing time and improve input accuracy. This technology can be applied to information collection systems that gather large amounts of documents from multiple information sources and build comprehensive databases.

[0111] Step 2: Analysis and summarization of legal amendment information (utilizing generative AI) The Legal Amendment Information Acquisition Department transmits uploaded or automatically acquired documents to the Generative AI Execution Department for parsing using natural language processing technology. The department generates prompts for the generative AI, such as "Summarize the following amendments to the Payment Services Act and extract key amendment points, implementation date, additional obligations, strengthening of penalties, and transitional measures," "Summarize the Executive Council's interpretation policy, practical considerations, and impact on electronic payment services from the following full text of guidelines," "Summarize key opinions and the Executive Council's responses from the following public opinion results," and "Extract hypothetical questions and answers from the following Executive Council FAQs," instructing the AI ​​to parse these prompts. The AI ​​automatically extracts and summarizes the main points of the amended law, amendment points, implementation period, new rules to be followed, impact on the industry, additional obligations, strengthening of penalties, and transitional measures, storing this information in the Database Management Department. Specifically, the Legal Amendment Information Acquisition Department transmits structured text data to the Generative AI Execution Department. The Generative AI Execution Department then activates an encoder-decoder type deep learning model. The data input to this model is a tensor combining extracted text data and a sequence of labeled instructions. The input data has a dimension of, for example, an 8192-dimensional context window. The elements of the input data are word embedding vectors and attention masks. As part of the preprocessing, this processing unit divides long texts into appropriate blocks according to the model's maximum input length. Specific examples of the input data include, firstly, blocks of data representing amendments to the financial settlement law; and secondly, text data from specific chapters of guidelines. This model uses a self-attention mechanism to calculate long-distance dependencies within the text. The output data from this model is a probability distribution of the labeled text that constitutes the summary. The output data is a set of summary vectors for each block. Each output value represents the importance score of the extracted legal elements. Specific examples of the output data include, firstly, date data representing the implementation date; secondly, text data representing additional obligations written in separate clauses; and thirdly, structured data representing changes to penalty provisions. This post-processing unit integrates the summary results from each block. This post-processing unit generates the final summary document. This post-processing unit stores the generated summary document in a predetermined table in a relational database. Through this configuration, the system parses lengthy legal documents at a speed far exceeding human reading comprehension. This system achieves the technical effect of preventing omissions in the extraction of important legal elements and providing comprehensive coverage of guarantees. This technology can be applied to use cases of legal due diligence involving the automatic extraction of risk clauses from large contracts or legal documents.

[0112] Step 3: Influencing specific departments (utilizing generative AI) The Department of Impact Assessment cross-references the business content, business processes, areas of responsibility, managed assets, external clients, related systems, data categories, existing regulations and manuals, educational materials, past legal amendment response records, internal audit findings, external audit reports, business contracts, system design documents, operation procedures, customer service manuals, franchisee management ledgers, transaction monitoring rules, risk assessment forms, internal notification records, and past administrative guidance response records of various departments, including service operations, system development, customer support, risk management, sales, finance, legal affairs, and audit. This data is then compared with the key points of legal amendments extracted by AI. The AI ​​is then fed a prompt such as, "Compare the following key points of legal amendments with the business content databases of each department and list the affected departments. Also extract highly relevant business processes and managed assets." The AI ​​analyzes the correlation between the business content of each department and the legal amendment content from multiple perspectives, automatically identifying the affected departments and prioritizing those with high impact. Specifically, this Department of Impact Assessment uses a graph neural network relational analysis algorithm. This processing unit constructs a knowledge graph with the business content and managed assets of each department as nodes and their dependencies as edges. The input data to this model consists of text vectors of key legal amendment points and the adjacency matrix of the knowledge graph. The dimensionality of this input data is a multidimensional tensor dependent on the number of nodes and the dimension of features. The elements of this input data are the attribute vectors of each node and the weights of the edges. As preprocessing, this processing unit uses a named entity recognition model to extract keywords from the document groups of each department, generating feature vectors for the nodes. Specific examples of this input data include, firstly, the vector data of API specifications managed by the system development department; and secondly, the vector data of response manuals from the customer support department. This model uses a message passing algorithm to propagate information between nodes. The output data from this model is an array of influence scores for each department's nodes. The output data is in the form of a list of floating-point numbers taking continuous values ​​between 0 and 1. Each output value represents the probability that a specific department is forced to make business changes due to legal amendments. Specific examples of this output data include, firstly, an influence score of 0.95 for the service operations department. Second, an impact score of 0.20 can be given for the finance department. This follow-up processing unit identifies departments with impact scores exceeding a predetermined threshold as affected departments. This follow-up processing unit sorts the departments in descending order of their scores, generating a priority list. Through this structure, the system not only achieves consistency in surface-level keywords but also enables advanced impact assessments that consider potential dependencies in business processes. This system effectively prevents specific omissions in the scope of impact and optimizes initial organizational responses. This technology can be applied to use cases of impact scope analysis tools during large-scale system changes or organizational restructuring.

[0113] Step 4: Automatic generation of solutions and data (utilizing generative AI) The Response Plan Generation Department uses generative AI to generate response plans and materials for each department identified as affected. The AI ​​provides prompts such as: "Based on the following key points of legal amendments and the department's business content, create necessary application process change plans, system specification change plans, customer notification plans, existing manual revision plans, new self-confirmation service proposals, franchisee review benchmark revision plans, fault response procedure reassessment plans, user compensation system implementation plans, FAQ update plans, educational material revision plans, internal audit checklist revision plans, external client response plans, API specification change plans, database design change plans, and risk assessment procedure reassessment plans," "Create briefing scripts and FAQ drafts based on the following content," and "Create customer notification templates based on the following content." Specifically, this Response Plan Generation Department drives a conditional text generation model. The data input to this model consists of a vector of configuration files for the affected departments, a vector of key points of legal amendments, and metadata defining the template structure of the materials to be generated. The dimension of this input data is a high-dimensional tensor combining multiple vectors. The elements of this input data include department-specific constraints and output format indicators. As part of the preprocessing, this processing unit retrieves response data from similar past cases and adds it as context to the input data. Specific examples of this input data include: first, prompt data indicating API specification changes for the system development department; second, instructions for generating customer notification templates for the sales department. This model uses a Transformer decoder to autoregressively generate text. The output data from this model consists of the generated response text and a structured task list. The output data is in Markdown format and JSON format task objects. Each output value represents a specific action item and its execution steps. Specific examples of this output data include: first, a draft SQL script describing a database schema change; second, the text of an email for customers; and third, an array of audit checklist items. This post-processing unit performs syntax checking on the generated text. This post-processing unit stores the generated data in dedicated directories for each department. Through this structure, the system instantly generates practical responses perfectly suited to the context of each department. This system dramatically improves business efficiency by significantly reducing the enormous time required to create data from scratch. This technology can be applied to use cases of automated planning and initiation tools in consulting or project management.

[0114] Step 5: Distribution, viewing, and editing of generated materials and response plans. Through the user interface and document viewing and editing departments, responsible personnel from various departments can access the server from their terminals to view and download generated summaries, lists of affected departments, response plans, notification materials, FAQs, briefing scripts, customer notification templates, educational materials, contract templates, audit checklists, external client response materials, API specifications, database design documents, risk assessment forms, etc. Responsible personnel can either directly generate materials using AI or add, delete, or modify materials by department. Specifically, this system utilizes a content delivery network to efficiently distribute generated materials to various client terminals. This processing department caches the generated data on edge servers. This processing department responds to user requests from the geographically nearest server. This processing department strictly verifies document viewing permissions based on a role-based access control model. This processing department blocks access requests from unauthorized users. This processing department applies optimistic concurrency control to document editing operations. This processing department prevents data inconsistencies when multiple users edit the same document simultaneously. This processing department assigns a digital signature to the edited document. This processing department cryptographically guarantees the authenticity and integrity of the document. Through this architecture, the system establishes a secure and high-speed information sharing foundation for large-scale organizations. This system distributes network bandwidth load. This system effectively prevents data tampering and achieves secure document management that meets compliance requirements. This technology can be applied to use cases of secure shared file systems for enterprises handling highly confidential design or financial data.

[0115] Step 6: Implementation of the briefing and notification activities Each department uses AI-generated briefing scripts and informational materials to conduct briefings, internal company training sessions, briefings for franchisees, briefings for external clients, and briefings for the group company.

[0116] For example, the service operations department holds internal briefings on the new personal verification process, the customer support department conducts training on new steps for customer response, and the sales department holds new service briefings for franchisees.

[0117] Specifically, this system starts the multimedia data generation module.

[0118] This processing unit inputs the automatically generated text data of the briefing script into the speech synthesis model.

[0119] This processing unit generates speech waveform data that mimics natural human speech from text data.

[0120] This processing unit automatically generates slide images for presentations from the text of the information and structured data.

[0121] This processing unit synchronizes the generated audio data with the slide images.

[0122] This processing unit encodes video files intended for on-demand distribution.

[0123] This processing unit will upload the generated video files to the streaming media server.

[0124] This processing department automatically sends the video URL to the responsible personnel in each department.

[0125] This processing unit embeds tracking code in the video player.

[0126] This processing department collects the start time, completion rate, and pause operation logs of each user's audio-visual content.

[0127] This structure eliminates the physical constraints and time costs associated with face-to-face briefings.

[0128] This system enables homogeneous and high-quality information delivery to be extended throughout the organization.

[0129] This system leverages the technical ability to quantitatively measure the penetration of informational activities through the analysis of audiovisual logs.

[0130] This technology can be applied to use cases such as e-learning platforms and compliance education systems within enterprises.

[0131] Step 7: Chatbot-based Q&A (Leveraging Generative AI) Through the chatbot response department, after a briefing or in daily business, when a person in charge enters a question from the terminal, the content is sent to the server. The generative AI on the server refers to the question content, key points of legal amendments, departmental business content, past Q&A history, etc., to generate the best answer and return it to the terminal.

[0132] As a sample prompt for generative AI, one could be: "Provide a practical and specific answer to the following questions, based on the key points of legal amendments and the department's business content."

[0133] Specifically, this chatbot response department functions as a conversational AI system with a conversation management mechanism.

[0134] This processing unit maintains each user's conversation history in an in-memory database.

[0135] The input data to this model consists of the current question text vector and the context vector of past dialogue history.

[0136] The dimension of this input data is a tensor that is dynamically expanded according to the number of dialogue rounds.

[0137] The elements of this input data include changes in user intent and information about the entities being referenced.

[0138] As part of this preprocessing, this processing unit extracts important keywords from the dialogue history and adjusts the weights of the input vector.

[0139] As a specific example of this input data, firstly, a question vector containing the pronoun "What are the details of this step?" can be cited.

[0140] Second, a vector of follow-up confirmation questions can be given in response to the previous answer.

[0141] This model uses the Transformer's cross-attention mechanism to calculate the relevance between the question and the context.

[0142] The output data from this model is a probability distribution of the response text that takes into account the context.

[0143] The output data is in the format of natural language strings.

[0144] Each output value represents a direct solution to the user's question.

[0145] As a specific example of this output data, the first example is the text data of the corresponding page that leads to a specific manual.

[0146] Second, textual data that provides step-by-step instructions for specific system operation procedures can be provided.

[0147] This follow-up processing department dynamically embeds URL links to internal company systems within the response text.

[0148] Through this structure, the system is not limited to a single question and answer session, but achieves continuous dialogue that maintains a complex context.

[0149] This system effectively improves users' self-resolution rate and reduces the workload of support departments.

[0150] This technology can be applied to operational support for complex business systems, as well as to use cases where internal help desks require specialized knowledge.

[0151] Step 8: Progress, History, and Notification Management The Log Management Department, Notification Management Department, and Historical Record Management Department work together to manage the system's full processing history, user operation history, AI output history, response progress, notification history, and past legal amendment response history in a unified manner.

[0152] The notification management department automatically publishes information on legal amendments and corresponding solutions, updated materials, briefing planners, FAQ updates, and reminders for unresolved tasks to all terminals. The historical record management department records the response content and progress of past legal amendments in chronological order and uses it as reference information for future legal amendments.

[0153] Specifically, this system uses an event-driven integrated management engine.

[0154] This processing unit aggregates log data sent asynchronously from various modules in the central message broker.

[0155] This processing unit uses a stream processing engine to parse data in real time.

[0156] This processing department monitors the status of tasks assigned to each department.

[0157] This processing department calculates the remaining time until the task deadline.

[0158] This processing unit automatically triggers an upgrade event when the remaining time is below a specified threshold.

[0159] Based on the escalation event, this processing department instructs the notification management department to generate a reminder message.

[0160] Within the Historical Records Management Department, this processing department submits the data from completed tasks to the time series database.

[0161] This department calculates the overall project schedule.

[0162] This processing department will format the calculated progress rate into data for chart plotting on the dashboard.

[0163] This structure allows the system to visualize the complex and intertwined dependencies and progress of numerous tasks.

[0164] The system's early detection and response delays enable proactive intervention.

[0165] This system effectively reduces project management costs and ensures reliable responses to legal amendments.

[0166] This technology can be applied to use cases such as large-scale system migration projects and progress management tools for company-wide compliance audit procedures.

[0167] (specific use case) For example, consider a scenario where a major revision to the fund settlement method is announced, requiring electronic settlement service providers to strictly enforce new personal verification procedures, add obligations to retain transaction records, change the reporting criteria for suspicious transactions, introduce a user compensation system, and strengthen security requirements when APIs are linked.

[0168] In this situation, the legal or risk management personnel upload amended provisions, guidelines, notices from the Executive Yuan, insights from industry groups, public opinions, FAQs from the Executive Yuan, and relevant news materials to the system from the terminal, or automatically obtain them from external legal databases or administrative agency websites through API linkage or crawling.

[0169] The Legal Amendment Information Acquisition Department receives these documents and generates prompts for the Generative AI Execution Department, such as "summarize the following amendments to the Payment Services Act and extract information on key amendments, implementation dates, additional obligations, strengthening of penalties, and transitional measures," "summarize the Executive Council's interpretation policy, practical considerations, and impact on electronic payment services from the following full text of guidelines," "summarize the main opinions and the Executive Council's responses from the following public opinion results," and "extract hypothetical questions and answers from the following Executive Council FAQs," instructing the AI ​​to analyze them.

[0170] Specifically, this system applies a natural language processing pipeline to a wide variety of input documents.

[0171] This processing department extracts text data from the document group and constructs a corpus.

[0172] This processing unit performs named entity recognition algorithms on the corpus.

[0173] The input data to this model is a sequence of word vectors containing texts with technical terms related to fund settlement algorithms.

[0174] The dimension of this input data is, for example, 1024 dimensions.

[0175] The elements of this input data are the part-of-speech information of each word and its contextual embedding representation.

[0176] As part of this preprocessing, this processing unit uses a custom dictionary specialized for the financial field to perform morpheme parsing.

[0177] As a specific example of this input data, the first example is a text vector containing the phrase "strictness of personal confirmation procedures".

[0178] Second, a text vector containing the phrase "security requirements when API is linked" can be cited as an example.

[0179] This model uses a bidirectional long short-term memory network and a conditional random field for sequence labeling.

[0180] The output data from this model is the probability distribution of entity labels for each word.

[0181] The output data is formatted according to the BIO format of the label arrangement.

[0182] Each output value indicates which category a particular word belongs to, such as "obligation", "penalty", or "date".

[0183] As a specific example of this output data, the first example is date data that has been labeled "Implementation Date".

[0184] Second, textual data of action items labeled "additional obligations" can be cited.

[0185] This post-processing unit will map the extracted entities into the schema of the relational database.

[0186] This system uses this structure to automatically build a structured database of legal requirements from unstructured documents.

[0187] This system effectively eliminates data input based on manual human intervention and dramatically improves the accuracy of information extraction.

[0188] This technology can be applied to use cases of regulatory technology solutions that monitor changes in financial regulations and automatically assess the impact on a company's internal systems.

[0189] Generative AI extracts key points such as "strengthening of personal confirmation procedures", "additional obligation to retain transaction records", "new AML / CFT requirements", "strengthening security benchmarks when API linkage is strengthened", "importing user compensation system", "changing reporting benchmarks for suspicious transactions", "strictening the review benchmarks for franchise stores", and "new reporting obligations when system failures are established", and stores them in the database management department.

[0190] The Department of Impact Assessment will compare the business content, business processes, areas of responsibility, managed assets, external clients, related systems, data categories, existing regulations and manuals, educational materials, past legal amendment response records, internal audit findings, external audit reports, business contracts, system design documents, operation procedures, customer service, legal affairs, and audit departments' databases with the legal amendment points extracted by AI. The department will then input the following prompt into the AI: "Compare the following legal amendment points with the business content databases of each department and list the affected departments. Also extract highly relevant business processes and managed assets."

[0191] Specifically, the Department of Impact Assessment activates a high-dimensional similarity calculation engine using a vector space model.

[0192] This processing department will input the extracted key points of each legal amendment into the sentence encoder.

[0193] This processing department converts the key points of legal amendments into dense semantic vectors.

[0194] This processing department also converts the vast collection of business documents stored in the databases of various departments into semantic vectors.

[0195] The input data to this model consists of vectors of key points of legal amendments and vector groups of business documents from various departments.

[0196] The dimension of this input data is, for example, a 512-dimensional floating-point array.

[0197] The elements of this input data are the coordinate values ​​of the latent space representing the semantic features of the text.

[0198] As part of this preprocessing, this processing unit uses the TF-IDF method to calculate the weights of important words and phrases in the document and corrects their vector representations.

[0199] As a specific example of this input data, the first key point vector that can be cited is "strengthening the security benchmark when API linkage".

[0200] Second, the document vector of the "External API Connection Instructions" from the system development department can be cited.

[0201] This model comprehensively calculates the cosine similarity between the key point vector and each document vector.

[0202] The output data from this model is a matrix of matching scores for each department.

[0203] The output data is in the format of a two-dimensional numerical matrix with departments as rows and key points of legal amendments as columns.

[0204] Each output value represents the strength of the impact on a specific department from a particular legal amendment.

[0205] As a specific example of this output data, the first example is the similarity score of 0.92 between the system development department and the enhanced security benchmark.

[0206] Secondly, the similarity score of 0.88 between the customer support department and the user compensation system can be cited as an example.

[0207] This subsequent processing unit analyzes the fraction matrix.

[0208] This subsequent processing department extracts pairs of departments and key points that have scores exceeding a threshold.

[0209] Based on the extracted pairs, this follow-up processing department automatically generates a list of affected departments and a mapping table of related business processes.

[0210] Through this structure, the system can instantly compare a large number of documents that are beyond human visual comprehension.

[0211] This system accurately identifies hidden dependencies and their scope of influence between specific departments.

[0212] This system effectively minimizes the risk of compliance violations due to omissions.

[0213] This technology can be applied to the use cases of repetitive business processes and impact analysis tools during large-scale corporate mergers or business integrations.

[0214] For example, if the service operations department is responsible for "personal confirmation", "franchise management", "settlement service application", "user compensation response", and "new service planning", AI will automatically identify it as a department affected by the strengthening of personal confirmation procedures and the introduction of user compensation system.

[0215] Given that the system development department is responsible for "settlement system development", "API management", "security response", "fault response", "database design", "log management" and "external system linkage", it is extracted as a department affected by the obligation to retain transaction records and changes in API specifications and enhanced security requirements.

[0216] As the customer support department is responsible for "customer response", "acceptance of personal confirmation documents", "FAQ response", "complaint handling", "acceptance of user compensation", and "guidance on personal confirmation procedures", it becomes a department affected by new obligations to explain to customers, changes in the personal confirmation process, and the user compensation system.

[0217] As the risk management department is responsible for "AML / CFT response", "transaction monitoring", "legal compliance", "risk assessment", "internal audit", "external audit response", "suspicious transaction reporting", and "development of risk reduction measures", it becomes the department affected by the new AML / CFT requirements and changes in monitoring benchmarks.

[0218] Specifically, the decision-making department of this impact department combines a large-scale language model based on the Transformer architecture with a hybrid inference engine based on graph neural networks.

[0219] This inference engine quantifies the hidden semantic relationships between the business content of each department and the key points of legal amendments in a high-dimensional vector space.

[0220] This system receives, as input data, the text of key points of legal amendments output from the Legal Amendment Information Acquisition Department and the text of business configuration files of various departments obtained from the database.

[0221] This system uses natural language processing techniques to tokenize these texts.

[0222] This system performs data cleaning, including the removal of unwanted stop words and stemming.

[0223] This system inputs the preprocessed labeled sequence into a multi-layered self-attention mechanism.

[0224] This system converts the input text into a dense vector of, for example, 768 dimensions.

[0225] As a specific example of input to this system, there are text vectors representing enhanced personal verification procedures and text vectors representing anti-money laundering requirements.

[0226] As another example of input to this system, there are text vectors representing the business content of the service operations department and text vectors representing the business content of the system development department.

[0227] This inference engine calculates the cosine similarity and Euclidean distance between the vector of key legal amendment points and the vector of departmental business operations.

[0228] This inference engine further uses graph neural networks to weight the inter-departmental dependencies within the organization.

[0229] For example, this inference engine represents the strength of data interaction between the system development department and the service operations department as the edge weights of the graph.

[0230] This inference engine integrates these computational results.

[0231] This inference engine, as output data, calculates an influence score representing the probability that each department will be affected by specific legal amendments.

[0232] This inference engine, as output data, generates influence category labels that classify the nature of the influence.

[0233] As a specific example of output from this system, there is an impact score of 0.95 for the service operations department and the category label "Application of Process Changes".

[0234] As another example of output from this system, there is an impact score of 0.88 for the system development department and the category label "System Modification".

[0235] As a subsequent processing step, the Department of Impact Assessment compares the calculated impact score with a pre-set threshold.

[0236] This impact department will designate departments whose scores exceed the threshold as impact departments.

[0237] The Department of Influence Determination will send the identifiers of the identified affected departments and relevant key data on legal amendments to the Response Plan Generation Department.

[0238] Even if the score of the affected department is below the threshold, departments that are close to the boundary value will be marked as departments requiring attention.

[0239] Through this structure, the system can extract with high precision implicit business relevance that is easily overlooked in keyword searches performed manually by humans.

[0240] This system will fully automate specific processes that affect departments.

[0241] This system significantly reduces the analysis time in the initial stage of responding to legal amendments.

[0242] This system effectively utilizes the computing resources of information processing devices to perform complex impact analyses at high speeds in large-scale organizations.

[0243] This technology is suitable for compliance management systems in enterprises with multiple business departments, such as financial institutions and telecommunications operators.

[0244] The Response Plan Generation Department, targeting departments identified as affected, used generative AI input to create necessary application process change plans, system specification change plans, customer notification plans, existing manual revision plans, new self-confirmation service proposals, franchisee review benchmark revision plans, fault response procedure reassessment plans, user compensation system implementation plans, FAQ update plans, educational material revision plans, internal audit checklist revision plans, external client response plans, API specification change plans, database design change plans, and risk assessment procedure reassessment plans. The department also created a briefing script and FAQ draft based on the following: The system automatically generates prompts such as "Customer Notification Templates," "Reassess the Personal Confirmation Process," "Create New User Instructions," "Revise Franchise Management Regulations," and "Develop Steps for Implementing a User Compensation System" for the service operations department; "Add Transaction Record Saving Function," "API Specification Changes," "Strengthen Security Requirements," and "Automatic Reporting Function for Installation Failures" for the system development department; "Develop New Steps for Accepting Personal Confirmation Documents," "Update FAQs," and "Create Customer Notification Templates" for the customer support department; and "Import New AML / CFT Monitoring Benchmarks," "Reassess Risk Assessment Steps," and "Update Internal Audit Checklists" for the risk management department.

[0245] Specifically, the solution generation unit is driven by the computation module of an autoregressive text generation model.

[0246] This computational module utilizes a dedicated neural network model that has fine-tuned the resume data using past legal amendments.

[0247] The response plan generation department uses this as input data to generate a structured prompt that integrates key legal amendment data received from the affected department's assessment department, the target department's business configuration files, and data from similar past cases.

[0248] The solution generation department should use prompting engineering techniques to clearly indicate the instructions for the model.

[0249] The solution generation department converts the structured prompt into a tag sequence.

[0250] The solution generation unit is input into this neural network model as an input sequence of up to 4096 labels.

[0251] As a specific input example to this model, there is a prompt text indicating a change in the specification of the application programming interface for the system development department.

[0252] As another example of input to this model, there is prompt text indicating the steps for importing anti-money laundering benchmarks for the risk management department.

[0253] This neural network model parses the context of the input labeled sequence through a multi-layered attention mechanism.

[0254] This neural network model predicts the next label sequentially based on a probability distribution.

[0255] This neural network model serves as output data, generating structured text data containing a list of specific tasks to be addressed.

[0256] As a specific example of the output from this model, there is a dataset containing the task name "Change of Authentication Token Specification", priority score "0.9", and estimated working time "40 hours".

[0257] As another example of output from this model, there is a dataset containing the task name "Production of Customer-Oriented Notifications", a priority score of "0.8", and an estimated working time of "15 hours".

[0258] The solution generation department should perform syntax parsing on the output structured text data as a subsequent processing step.

[0259] The solution generation department will register the parsing results as a new record in the task management database.

[0260] The solution generation department generates drawing commands to display registered task data on the user interface of each department.

[0261] This solution generation department analyzes the dependencies between generated tasks and automatically calculates the critical path.

[0262] Through this structure, the system does not simply generate general text, but automatically generates executable task plans that conform to the company's past response experience and actual business situation.

[0263] This system suppresses the illusions created by generative models.

[0264] This system reduces the human resources required to develop response plans.

[0265] This system improves the accuracy of project management by quantitatively indicating task priorities and timeframes.

[0266] This technology is extremely useful as a project schedule management tool for company-wide legal modification projects led by the legal or compliance departments.

[0267] AI automatically generates legal amendment notices, FAQ drafts, briefing scripts, customer notification templates, educational materials, revised user manuals, revised contract templates, audit checklists, external client response materials, API specifications, database design documents, risk assessment forms, etc. for various departments. Each responsible person can view, edit, and download these materials through the user interface department and the document viewing and editing department.

[0268] For example, the service operations department can add descriptions of the company's unique applications to the AI-generated personal confirmation process plan; the system development department can add the company's inherent system requirements to the API specification change plan; the customer support department can add frequently asked questions to the FAQ draft; and the sales department can add the company's advantages to the new service proposal materials.

[0269] Specifically, this document reading and editing department has a data management module with a distributed version control algorithm installed.

[0270] This data management module stores the generated document data as tree-structured data objects in the memory.

[0271] This user interface unit receives editing operation events sent from the operator's terminal in real time.

[0272] This document, as input data, is used by the reading and editing department to obtain a differential data array representing the additions, deletions, and changes made to the text by the responsible party.

[0273] As a specific example of input to this system, there exists string data representing additional descriptions of specific authentication parameters in the application programming interface specification.

[0274] As another example of input to this system, there is additional string data representing the answers to specific error codes in a draft of common questions.

[0275] This document was prepared and edited by the editorial department using comparison algorithms such as the Myers difference algorithm to parse and obtain the difference data array.

[0276] The document review and editing department clearly distinguishes and records the original text generated by the artificial intelligence model from the text edited by human authors.

[0277] The editorial department uses natural language processing to categorize the intent behind the edited content.

[0278] This document, viewed and edited by the editorial department, serves as output data, generating complete text data of the updated document, a unique version hash value, and metadata from the editorial department.

[0279] As a specific example of output from this system, there exists a meta-dataset containing the updated instruction manual text, the hash value "a1b2c3d4", and the editor identifier.

[0280] As another example of output from this system, there exists a metadata set containing the updated question and answer text, the hash value "e5f6g7h8", and the timestamp.

[0281] As part of the subsequent processing by the review and editorial department, the extracted differential data and metadata will be transferred to the learning data repository.

[0282] This system will reuse the accumulated editing history data as teacher data for reinforcement learning and suggestion optimization of generative artificial intelligence models.

[0283] This system automatically adjusts the weighting parameters for the next document generation for areas that are frequently modified.

[0284] Through this structure, the system efficiently restores the tacit knowledge possessed by human experts and the inherent requirements of the site into the system's internal knowledge base.

[0285] This system forms a feedback loop that continuously improves the accuracy of model generation.

[0286] This system saves storage capacity by only saving differential data.

[0287] This system effectively prevents conflicts during collaborative editing by multiple people and ensures data consistency.

[0288] This technology proves highly useful in collaborative work platforms where multiple departments work together to create and review documents.

[0289] AI-generated materials are also used in briefings and internal company training sessions, briefings for franchisees, briefings for external clients, and briefings for the group company.

[0290] For example, the service operations department held an internal briefing on the new personal verification process, the customer support department conducted training on the new steps for customer response, and the sales department held a new service briefing for franchisees.

[0291] After a briefing or during routine business operations, when a person in charge enters a question through the terminal, the chatbot response department and the generative AI on the server provide practical and specific answers to questions such as "What are the specific contents of the new personal confirmation requirements?", "Under what circumstances does the obligation to retain transaction records apply?", "When should customers be notified?", "What are the specific contents of the API specification change?", "What is the basis for adding audit items?", "What is the scope of the user compensation system?", "How should we deal with external clients?", and "What are the legal considerations for expanding into new markets?"

[0292] Specifically, the chatbot's response unit functions as an advanced question-answering system employing a retrieval-enhanced generation architecture.

[0293] This system has a vector database that stores company internal regulations, legal amendment information, and past response history as multi-dimensional vectors.

[0294] The chatbot's response unit receives natural language-based question text strings sent from the operator's terminal as input data.

[0295] The chatbot's response unit uses an embedding model to convert the question text string into a dense feature vector of, for example, 1024 dimensions.

[0296] As a specific input example to this system, there exists a feature vector derived from the question text "What are the specific contents of the changes to the application programming interface specification?".

[0297] As another example of input to this system, there exists a feature vector derived from the question text "What is the basis for adding audit items?"

[0298] The chatbot's response department uses the generated question vectors to search within a vector database.

[0299] The chatbot's response unit extracts blocks from multiple documents that are most relevant to the question based on a cosine similarity equidistant scale.

[0300] The chatbot's response department combines the extracted document blocks with the original question text to construct input prompts for the generative artificial intelligence model.

[0301] This generative artificial intelligence model infers answers based on the provided context.

[0302] The chatbot's response unit generates, as output data, a natural language-based response text, an identifier referencing the source document, and a probability score indicating the confidence level of the response.

[0303] As a specific example of output from this system, there is a response text explaining the details of the specification change, a reference source "System Development Guide, 3rd Edition", and a confidence score of 0.92.

[0304] As another example of output from this system, there is a response text explaining the point of change in the audit baseline, the reference source "Internal Audit Manual Revision", and a confidence score of 0.88.

[0305] The chatbot response unit evaluates the calculated confidence score as a follow-up process.

[0306] If the confidence score is below a preset threshold, the chatbot's response department will retain the sending of the reply text.

[0307] This chatbot's response department automatically generates upgrade requests for human experts and notifies the relevant departments.

[0308] The chatbot's response department collects user feedback on the answers and dynamically updates the weights of the retrieval algorithm.

[0309] Through this structure, the system prevents fact misidentification that may occur in a large-scale language model alone.

[0310] This system provides accurate answers in real time, always based on the latest internal company information.

[0311] This system significantly reduces the time required for information retrieval by those in charge.

[0312] This system effectively reduces the workload of consulting services and improves the overall operational efficiency of the organization.

[0313] This technology is best suited for help desk systems and internal corporate portals that handle complex internal rules and frequently changing legal information.

[0314] Furthermore, the Log Management Department, Notification Management Department, and Historical Record Management Department work together to uniformly manage the system's full processing history, user operation history, AI output history, response progress, notification history, and past legal amendment response history. The Notification Management Department automatically sends notifications such as "New legal amendment information added," "This department has been selected as an affected department," "New response plan generated," "Guide to holding briefings," "FAQ updated," "Unresolved tasks exist," "Response to external clients incomplete," and "Document distribution to the group company completed," as well as reminders for unresolved tasks. The Historical Record Management Department records the response content and progress of past legal amendments in chronological order, which can be used as reference information for future legal amendments. Specifically, the Log Management Department, Notification Management Department, and Historical Record Management Department communicate with each other through an asynchronous messaging system based on an event-driven architecture. The Log Management Department continuously writes all events occurring within the system as timestamped structured log data into a time-series database. The Historical Record Management Department executes a parsing engine that includes machine learning-based anomaly detection models and task delay prediction models. This analysis engine takes as input data a feature vector representing the progress status of tasks, extracted from a time-series database. A specific example of input for this system includes a numerical vector containing the elapsed time since a particular task was created, the frequency of logins by the responsible party, and the number of updates to related documents. Another example of input includes numerical data representing the difference between the average completion time of similar past tasks and the current elapsed time. This analysis engine processes this input data using a recurrent neural network or gradient boosting decision tree algorithm. This analysis engine quantifies the risk of each task not being completed within the deadline. As output data, this analysis engine generates a delay probability score and a corresponding alert level classification for each task. A specific example of output from this system is a delay probability score of 0.85 and an alert level of "High" for task identifier "TSK-1024". Another example of output from this system is a delay probability score of 0.42 and an alert level of "Low" for task identifier "TSK-2048". This notification management department monitors the output delay probability scores for subsequent processing. When the notification management department detects that a score exceeds a predetermined threshold, it automatically generates and sends push notifications urging immediate action to the terminals of the person responsible for the task and the manager. This department tracks the open status of notifications and subsequent task progress, dynamically adjusting the re-notification schedule as needed. The history management department aggregates log data from completed tasks and calculates the response performance indicators for each department. Through this structure, the system goes beyond simple information recording and standardized reminders; it proactively predicts project delay risks using a data-driven approach. This system enables managers to take countermeasures before problems occur.This system minimizes the risk of compliance violations in projects where deadlines are strictly regulated by legal amendments. It provides visualization of the overall system operation and optimizes resource allocation. This technology plays a crucial role in enterprise-oriented project management systems that manage large-scale and complex workflows.

[0315] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires voice representing the user's input on the result of the specific processing. The control unit 46A sends the voice data representing the user's input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0316] Data generation model 58 is what is known as generative AI (Artificial Intelligence).

[0317] One example of data generation model 58 includes ChatGPT (registered trademark) (Internet search).<URL: https: / / openai.com / blog / chatgpt> Generative AI, etc.

[0318] The data generation model 58 is obtained by performing deep learning on a neural network. A prompt containing instructions is input to the data generation model 58, along with inference data such as speech data representing speech, text data representing text, and image data representing images (e.g., data from still images or moving images). The data generation model 58 infers the input inference data based on the instructions shown in the prompt and outputs the inference result in one or more data forms, such as speech data, text data, and image data. The data generation model 58 may include, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization. The specific processing unit 290 performs the aforementioned specific processing while using the data generation model 58. The data generation model 58 may also be a fine-tuned model to output inference results from prompts without instructions; in this case, the data generation model 58 can output inference results from prompts without instructions. The data processing device 12, etc., includes multiple data generation models 58, including AI other than generative AI. AI other than generative AI includes, but is not limited to, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes. Furthermore, AI can also act as an AI agent. Additionally, when the processing described above is performed by AI, this processing may be partially or entirely performed by AI, but is not limited to this example. Moreover, processing performed by AI, including generative AI, can be replaced by rule-based processing, and vice versa.

[0319] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart device 14, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart device 14. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart device 14 or external devices, and the smart device 14 acquires or collects information required for processing from the data processing device 12 or external devices.

[0320] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart device 14.

[0321] [Second Implementation] Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0322] like Figure 3 As shown, the data processing system 210 includes a data processing device 12 and smart glasses 214. One example of the data processing device 12 is a server.

[0323] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0324] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, and communication I / F 44 are also connected to the bus 52.

[0325] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0326] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0327] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0328] Figure 4 An example of the main functions of the data processing device 12 and the smart glasses 214 is shown. Figure 4As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0329] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0330] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0331] In the smart glasses 214, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart glasses 214 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290.

[0332] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart glasses 214. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0333] (Example 1) Since the specific processing flow is the same as that in Embodiment 1 described in the first embodiment above, the description is omitted.

[0334] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0335] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0336] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart glasses 214, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart glasses 214. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart glasses 214 or external devices, and the smart glasses 214 acquires or collects information required for processing from the data processing device 12 or external devices.

[0337] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart glasses 214.

[0338] [Third Implementation] Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0339] like Figure 5 As shown, the data processing system 310 includes a data processing device 12 and a head-mounted terminal 314. An example of the data processing device 12 is a server.

[0340] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0341] The head-mounted terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, display 343, and communication I / F 44 are also connected to the bus 52.

[0342] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0343] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0344] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 a...

Claims

1. A system for supporting responses to legal amendments, comprising a legal amendment information acquisition unit, an affected department determination unit, and a response plan generation unit, characterized in that, The Legal Amendment Information Acquisition Department uses generative AI to automatically parse multiple legal amendment-related documents, such as amended legal provisions, guidelines, public opinions, and legislative discussion records, and extracts and summarizes the main points, amendments, implementation period, new rules to be followed, and impact on the industry. The Affected Department Determination Department uses generative AI to compare the business content database of multiple departments registered within the company with the key points of legal amendments extracted by the Legal Amendment Information Acquisition Department. It compares the relevance of each department's business content to the legal amendment content by keywords or business processes, automatically determines the departments affected by the legal amendments, and prioritizes departments with high impact. The Response Solution Generation Department uses generative AI to propose application change plans, data revision plans, and amendments to existing application manuals and internal company regulations for each department determined to be affected. It also proposes new business opportunities that may arise due to the legal amendments. In addition, it automatically generates notification materials, draft FAQs, briefing scripts, and time allocation plans for each department regarding the legal amendments, and provides real-time answers to questions after the briefings by a generative AI-based chatbot.

2. The system as claimed in claim 1, wherein, The response generation department is equipped with the ability to analyze the application and management content implemented in the affected departments through generative AI, specifically suggest how to modify which data and application processes, automatically generate revised application instructions, and also suggest the possibility of new business arising from legal changes by department.

3. The system as described in claim 1, wherein, The legal amendment information acquisition department is equipped with the ability to automatically analyze multiple legal amendment-related documents, such as amended legal provisions, guidelines, public opinions, and discussion records during the legislative process, through generative AI. It can extract and summarize the main points of the amendment, the points of modification, the implementation period, the new rules to be followed, and the impact on the industry. The summary results are automatically linked to the impact department determination department and the response plan generation department.

Images