Data processing system

Abstract

The present application relates to a system that uses generative AI to automatically and multi-angle evaluate new ideas submitted by employees, integrates business performance data to calculate comprehensive contribution, visualizes the evaluation results on dashboards and the like, automatically designs the best rewards or recognition content through generative AI, and automatically generates and distributes feedback or reports such as evaluation results or improvement suggestions, progress status, business impact prediction, and the like, characterized in that the fairness, transparency and efficiency of idea evaluation are greatly improved, and the promotion of innovation and retention of outstanding talents in the organization are achieved.

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 addresses is that, within enterprises or organizations, the evaluation of employees' new ideas is often influenced by the evaluator's subjectivity, interpersonal relationships, and organizational dynamics. This leads to a lack of proper recognition of innovative ideas and talented individuals, resulting in the loss of capable employees or stagnation of innovation. Furthermore, the separation of creative evaluation from routine performance evaluation leads to an inadequate understanding of employees' overall contributions, and the design of compensation or recognition becomes uniform and inefficient, reducing employee motivation or hindering the growth of the entire organization. In addition, the reliance on individuals for evaluation, feedback, and progress reporting, coupled with inefficiency, compromises the transparency and fairness of evaluations, and fragments communication between employees and superiors, hindering the comprehensive achievement of promoting innovation and utilizing talent within the organization. To solve these problems, this invention aims to automate and standardize creative evaluation, integrate it with business performance evaluation, design personalized compensation, and automate continuous feedback and progress reporting by utilizing generative AI. By providing a fair and transparent mechanism for evaluating employees' creative contributions, this invention promotes innovation, retains top talent, and drives the growth of the entire organization. Summary of the Invention

[0005] The means to solve the problem involved in this invention is to provide a creative evaluation system using generative AI. The system includes: a creative receiving unit that receives and records the content, submitter information, submission date and time, etc., of new creative ideas submitted by employees in a database; an automatic evaluation index calculation unit that uses generative AI to automatically calculate multiple evaluation indicators such as the innovativeness, feasibility, and economic impact of the received creative ideas through natural language processing or knowledge base analysis; a business performance integration unit that integrates the employee's general business performance data with the evaluation indicators to calculate a comprehensive contribution; and a visualization display unit that displays the evaluation indicators and comprehensive contribution in charts or tables, etc. The results are displayed visually on the dashboard; the Compensation Design Department uses generative AI to automatically design the best compensation or recognition for each employee based on the overall contribution; the Feedback Generation Department uses generative AI to automatically generate evaluation results or improvement suggestions and provide them to employees or supervisors; and the Progress Reporting Department uses generative AI to automatically report the evaluation status or progress of ideas, the predicted impact on business, etc., to relevant personnel. The goal is to achieve automation and standardization of idea evaluation, integrated evaluation with business performance, personalized compensation design, and continuous feedback and automated progress reporting through the coordinated operation of these departments, providing a fair and transparent mechanism for evaluating employees' creative contributions. 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 8This 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 specific and detailed manner. Hereinafter, for each constituent element, the role division between the server and the terminal, the method of utilizing generative AI, specific processing examples, variations in application, and envisioned application examples will be described in a comprehensive manner.

[0040] First, the creative receiving department operates via a dedicated input screen on the terminal side. Employees can input creative titles, detailed descriptions, related materials (such as PDFs, images, videos, design drafts, reference URLs, etc.), submitter information (employee ID, department, position, contact information, etc.), submission date and time, related project names, envisioned effects, and topics from terminals such as laptops, tablets, and smartphones. For example, diverse creative ideas of various fields and scales can be input, such as "a proposal for an AI-based automated FAQ response system within the company," "an SNS campaign plan for acquiring new customers," and "a delivery route optimization algorithm for reducing logistics costs." Data input on the terminal side is instantly sent via encrypted communication and recorded in a database on the server side. Furthermore, as input assistance functions, the terminal side can also be equipped with features such as automatic retrieval and display of similar previously submitted creative ideas, automatic saving of input content, a preview function before submission, and editing and appending functions after submission. Specifically, this receiving department acquires the diverse data sent from the terminal side's input interface as structured data. This receiving department encodes the input text data using a specified character encoding format. This receiving unit inputs image data, which is attached as related information, into a feature extractor using a convolutional neural network. This feature extractor converts the image data into, for example, a 256-dimensional feature vector. This receiving unit then sends this structured data and feature vector to the server via a secure communication protocol. As an input assistance function, this receiving unit deploys a lightweight natural language processing model on the terminal's memory. This model analyzes the text stream being input by the user in real time. This model infers the probability distribution of the next word to appear. This receiving unit displays the word with the highest probability as a candidate for auto-completion on the screen. Furthermore, this receiving unit converts the input text into word embedding vectors. This receiving unit calculates the cosine similarity between the vector group and past creative ideas in the database. This receiving unit extracts past creative ideas with similarity exceeding a predetermined threshold and displays them on the screen as reference information. This system performs these preprocessing steps on the terminal side. Therefore, this system reduces communication traffic to the server side. Simultaneously, by standardizing the input data, this system drastically reduces the computational load of subsequent analysis and processing by artificial intelligence models. This feature can reduce the user's input burden and enable high-quality data collection in use cases such as patent proposal systems in manufacturing and new business proposal systems in information and communication enterprises.

[0041] The server-side evaluation metric calculation unit consists of an advanced natural language processing engine utilizing generative AI, a knowledge base comparison module, and scoring algorithms. For example, the generative AI analyzes the detailed descriptions of submitted ideas, performing keyword extraction, contextual understanding, logical structure analysis, and automatic identification of relevant technologies. The generative AI interacts with various knowledge bases within the company, including past idea databases, patent databases, academic paper databases, industry news, publicly available information from competitors, and internal and external technical documents, via API to evaluate the innovativeness of submitted ideas from multiple perspectives. For example, it automatically determines "which part is novel compared to the existing internal FAQ system," "whether the same technology is already widespread in the industry," and "whether it overlaps with past internal proposals." In assessing feasibility, the generative AI also references internal technical resources (such as available developers, budget, compatibility with existing systems, past project success rates, and the company's expert network), external technological trends, and legal and regulatory information to automatically extract the necessary conditions and risk factors for implementation and assign them scores. In economic impact assessment, generative AI uses market size forecasting data, company financial performance, industry average cost structure, revenue data from similar past cases, ROI simulation, etc., to automatically calculate various economic indicators such as increased sales, cost reduction, brand value enhancement, and improved customer satisfaction. For example, in the case of an "SNS campaign plan," it refers to the effectiveness data of past SNS initiatives and cases from competitors to predict the expected reach, number of customers acquired, cost-effectiveness, etc. These evaluation indicators are automatically weighted by generative AI and calculated as a comprehensive evaluation score. The evaluation process can also be implemented multiple times, including an initial evaluation after submission, a re-evaluation after a certain period, and phased evaluations based on project progress. Specifically, this computing unit is constructed around a large-scale language model with a self-attention mechanism. This computing unit converts the input text data into an array of integer identifiers through a tokenizer. This computing unit assigns positional encoding to this array, generating, for example, a 768-dimensional input tensor. This computing unit inputs this input tensor into the large-scale language model. Specific examples of input data include text vectors representing differences from existing systems, time-series arrays of market size, and structured data on company resources. This large-scale language model optimizes weights through pre-training and fine-tuning using a cross-entropy error function. It performs a non-linear transformation of the input tensor in a high-dimensional space, inferring scores from the output nodes corresponding to each evaluation metric. The output data from this large-scale language model consists of scores of floating-point numbers normalized to the range of 0.0 to 1.0, and probability distributions representing the attention weights. Specific examples of output data include an innovation score of 0.85, a likelihood of realization score of 0.60, and an economic impact score of 0.75.This calculation unit combines the output scores as feature vectors. The combined feature vectors are then input into a multilayer perceptron to calculate the final comprehensive evaluation score. If the calculated comprehensive evaluation score exceeds a predetermined threshold, the calculation unit performs subsequent processing, automatically assigning an initial review approval flag to the idea. This system achieves high-precision semantic similarity determination based on vector operations in high-dimensional space. Therefore, this system effectively eliminates the bias of subjective human evaluations and significantly shortens evaluation time. This function is used in use cases such as company-wide business improvement competitions and new topic exploration in research and development departments to instantly filter large numbers of proposals.

[0042] The Business Performance Integration Department manages and integrates diverse performance data related to employees' routine business activities on the server side. This includes automatically collecting various data sources such as business evaluation scores (regular evaluations by supervisors or colleagues), project achievement (delivery deadlines, deliverable quality, customer feedback), attendance data (attendance rate, overtime hours, leave status), customer satisfaction scores, internal and external awards, and skills training records via API linkage or batch processing. Generative AI integrates this performance data with creative evaluation indicators, such as weighting "creativity evaluation score 70, business evaluation score 80, project achievement 90, customer satisfaction 85" to calculate overall contribution. The weighting model can be customized for each organization, automatically adjusted based on evaluation criteria such as "innovation-oriented" or "business stability-oriented." Furthermore, the generative AI learns employees' past contribution tendencies and growth curves, enabling future potential evaluation and automatic extraction of promotion candidates. Specifically, this integration department has a preprocessing module that maps heterogeneous data formats to a common multi-dimensional vector space. This integration department converts data collected from various systems into, for example, standardized 128-dimensional numerical vectors or time-series arrays representing performance trends over the past 12 months. These vectors and arrays are then input into a Long Short-Term Memory (LSTM) network or a one-dimensional convolutional neural network specifically designed for time-series analysis. Specific examples of input data include scalar values ​​representing performance evaluation scores and time-series arrays representing attendance records. The network extracts hidden features from the input time-series data to predict future performance. The output data from this network consists of floating-point scores representing overall contribution and probability distributions representing future potential. Specific examples of output data include an overall contribution score of 0.88, a promotion probability of 0.82, and a turnover risk of 0.05. The integration department stores the calculated scores and probability distributions in corresponding records in a database. This integration department then transmits these output data as input tensors to the subsequent compensation design department. This system improves the accuracy of multi-faceted evaluations through the standardization and integration of heterogeneous data. Simultaneously, the system leverages time-series analysis using machine learning models to reduce errors in future predictions. This feature enables data-driven, objective talent evaluation in use cases such as talent management systems in human resources departments and early identification programs for next-generation leaders.

[0043] The visualization department displays the evaluation and performance data aggregated on the server side in various formats on the dashboard screen on the terminal side. For example, it provides various visualization modes such as creative evaluation trend charts for each employee, creative submission rankings by department, heatmaps of overall contribution distribution, radar charts for each creative submission by evaluation item, Gantt charts of project progress, and analysis charts of the organization's overall innovation tendency. On the terminal side, in addition to employees being able to view their own evaluation status, past submission history, and feedback history, superiors and managers can have a comprehensive overview of their subordinates and the entire organization. For example, it can extract and display with one click "departments with the most creative submissions this month," "creations that received the highest evaluations in the past six months," and "employees with rapidly rising evaluation scores." Specifically, this visualization department has a dimensionality reduction engine that processes the high-dimensional evaluation vectors calculated on the server side. This engine uses principal component analysis algorithms or t-distribution random neighborhood embedding algorithms to reduce high-dimensional data to 2D or 3D coordinate data. This visualization department sends the compressed coordinate data to the terminal side in a lightweight data exchange format. This visualization department directly controls the rendering pipeline of the graphics processing unit on the terminal side. This display unit dynamically generates complex graphs such as heatmaps and scatter plots based on received coordinate data. The system employs an architecture that separates high-level computational processing on the server side from hardware-accelerated rendering on the terminal side. This makes visually understanding high-dimensional data easier. Simultaneously, the system leverages optimized rendering processing on the terminal side, completely preventing screen lag and stuttering when displaying large amounts of data points. This functionality provides an intuitive data analysis environment in use cases such as company-wide innovation portfolio monitoring dashboards for management and resource optimization simulators for department heads.

[0044] The compensation design department utilizes generative AI on the server side. This generative AI performs multivariate analysis on factors such as overall contribution, past recognition history, employee motivation, organizational compensation policy, budget status, employee career aspirations, and past incentive responses, automatically designing optimized compensation and recognition content for each employee. For example, based on organizational culture and employee aspirations, it proposes various incentives, including special recognition, monetary bonuses, stock options, promotion recommendations, new project leader appointments, overseas training opportunities, invitations to company events, access to skills enhancement training courses, introductions in company promotions, and company-wide points rewards. For instance, for employees who have historically prioritized career advancement over monetary rewards, promotion recommendations and project leader appointments are prioritized. The rationale and expected effects of the compensation design are also automatically generated by the generative AI as explanatory text and displayed to employees and supervisors on the terminal. Specifically, this design department employs a recommendation model based on a reinforcement learning algorithm using Markov decision processes. This recommendation model defines the employee's personal data as the state space and the combination of compensation plans as the action space. This recommendation model sets the employee performance improvement rate as the reward function and explores the optimal strategy using a Q-learning algorithm. The input data to this recommendation model consists of structured data such as comprehensive contribution, motivational tendency vector, and budget constraint matrix. Specific examples of input data include a contribution score of 0.9, a career aspiration category identifier of 3, and a remaining budget vector. The output data from this recommendation model consists of a one-hot vector representing the optimal combination of reward plans, and the expected effect score of each plan. Specific examples of output data include a special bonus probability of 0.95 and an overseas training recommendation probability of 0.80. This design department decodes the output one-hot vector. Based on the decoded information, this design department generates text data containing specific reward content and sends it to the visualization department for further processing. This system significantly reduces the computation time for large combinatorial optimization problems. Therefore, this system dynamically suggests rewards perfectly aligned with individual aspirations, maximizing the effect of employee motivation improvement. This functionality is applied in use cases such as personalized personnel systems in global enterprises promoting diverse work styles and startups requiring flexible incentive designs.

[0045] The Feedback Generation Department uses generative AI on the server side to automatically generate evaluation results and improvement suggestions. For example, when an idea is highly innovative but has a low feasibility, it suggests specific action plans such as "To solve technical issues, we recommend collaborating with experts within the company" or "Conducting additional market research can improve feasibility." Conversely, when the feasibility is high but the economic impact is limited, it generates practical suggestions such as "To maximize cost reduction, explore collaboration with other departments" or "As an additional proposal to existing customers, we recommend pilot implementation." Feedback is distributed to employees and supervisors through multiple channels, including dashboards on the client side, email notifications, and chatbot integration. Furthermore, the feedback content learns from employees' past reactions and improvement records, evolving into more personalized content. Specifically, this generation department has a text generation model using a decoder-only Transformer architecture. This generation model is pre-tuned using a dataset of excellent feedback cases within the company. The input data to this generation model consists of an evaluation score vector, text embedding vectors of past feedback records, and an array of labels representing improvement points. Specific examples of input data include an innovation score of 0.8, a feasibility score of 0.5, and past point vectors. This generative model uses a cluster search algorithm to explore the most natural and effective word sequences within the context to construct the article. The output data from this generative model is feedback text based on natural language and category identifiers for action plans. Specific examples of output data include text articles recommending market research and category identifiers for technical collaboration. The generation department asynchronously pushes and distributes the generated text to the notification module on the terminal side. The generation department performs subsequent processing based on the output category identifiers, automatically assigning contact information for relevant departments and links to reference documents. This system automatically generates personalized, high-quality feedback in real time. As a result, this system significantly reduces communication costs for management positions. Simultaneously, this system leverages the technical effect of high-precision, context-aware article generation based on natural language processing to homogenize the quality of proposals. This function helps guide business efficiency in use cases such as automated support for code reviews in large-scale development organizations and automated proposal grading systems in sales departments.

[0046] The progress reporting department uses generative AI on the server side to analyze the evaluation status and progress of creative ideas, as well as their impact on business predictions, automatically generating reports that all relevant personnel can view in real time. For example, it generates detailed reports including progress and impact predictions, such as "The new service proposal is currently in the empirical testing phase, and we expect increased sales in the initial evaluation," or "The business process automation solution requires further discussion due to remaining technical issues," and distributes these reports to relevant personnel through dashboards and email notifications on the terminal side. Furthermore, the generative AI analyzes past evaluation data and performance, automatically calculating the business impact predictions of creative ideas and their contribution to future organizational growth, such as providing specific numerical predictions like "If this creative idea is implemented, it is expected to reduce annual costs by 10%" or "There is a possibility of a 20% increase in new customer acquisition." Specifically, this reporting department has a prediction engine that uses graph neural networks. This prediction engine models the complex dependencies between projects as a graph structure of nodes and edges. The input data to this prediction engine consists of time-series tensors representing the progress status of projects and performance data from similar past projects. Specific examples of input data include arrays of task completion numbers over the past four weeks and vectors of member operation rates. This prediction engine performs graph convolution operations, updating the feature values ​​of each node while using a regression model to predict future values. The output data from this prediction engine includes a probability distribution of the scheduled completion date, continuous values ​​representing predicted cost reductions, and structured data for report construction. Specific examples of output data include a 0.15 probability of exceeding the scheduled completion date and a predicted cost reduction of 1.5 million yen. The reporting department injects the output predicted values ​​and structured data into a predefined template engine. The reporting department then performs subsequent processing, dynamically rendering interactive reports in Hypertext Markup Language format and distributing them to relevant personnel's terminals. This system significantly improves the accuracy of progress predictions for projects with complex dependencies. As a result, this system enables rapid decision-making based on real-time data updates. Simultaneously, this system achieves the technical effect of fully automating report creation and reducing manpower hours. This function provides data-driven, objective future predictions in use cases such as project management offices encompassing multiple agile development teams and monitoring return on investment in business planning departments.

[0047] Furthermore, for example, after an employee submits an idea from the terminal, the generative AI immediately performs an initial evaluation, maintaining employee motivation by returning a provisional score and simple feedback. For instance, the generative AI also handles the scheduling of regular review meetings, automatically adjusting the schedules of employees and supervisors, and automatically registering them in the terminal's calendar. For example, meta-analysis analyzing the organization's overall innovation trends, such as which departments are most creatively active and which types of ideas are more likely to receive high evaluations, is also performed by the generative AI on the server side, and can be presented to managers as reports on the terminal. For instance, when an employee submits multiple ideas, visualizing the evaluation progress and feedback history of each idea in a time series allows employees to experience a tangible sense of self-growth. Specifically, to achieve real-time evaluation, this system configures a lightweight, inference-specific neural network model on an edge server or in-memory cache. This evaluation module performs the vectorization of the input text and the inner product calculation of the learned weight matrix in milliseconds, returning a score immediately. In the scheduling function, this system applies a constraint satisfaction problem-solving algorithm. This algorithm takes the idle time matrix of multiple users as input and rapidly extracts the optimal solution from the exploration space that satisfies constraints such as meeting room availability and priority. In its meta-analysis function, the system executes non-hierarchical clustering algorithms such as K-means. The algorithm groups the feature vectors of highly rated ideas in a high-dimensional space, automatically extracting common trends and keywords. This system parallelizes these processes. Therefore, the system achieves a decentralized load reduction on the main server based on inference processing. Simultaneously, the system effectively reduces the computational cost of scheduling adjustments performed by optimization algorithms and improves the overall system responsiveness. This functionality minimizes application costs in use cases such as the operation of maker marathons in large enterprises with tens of thousands of employees and the automatic scheduling of daily one-on-one meetings.

[0048] Furthermore, for example, to cope with the multilingual environment within the company, generative AI automatically translates submitted ideas and feedback, and can also smoothly share and evaluate ideas among global branches. For example, to ensure employee anonymity, generative AI can be installed to mask submitter information and eliminate bias in evaluation. For example, when external experts and customers participate in the evaluation process, generative AI automatically adjusts the evaluation criteria and feedback content, and can also provide an interface for external evaluators to the terminal. Specifically, this system installs an encoder-decoder type neural machine translation model for multilingual support. This translation module maps the input text in any language to a common latent space vector. This translation module decodes the vector into text in the specified target language. In the anonymization processing unit, the named entity extraction model is activated. This anonymization processing unit specifically represents the tags such as personal names, department names, project names, etc. in the text. This anonymization processing unit performs preprocessing to replace the specific tags with specific mask symbols, completely eliminating information that can identify individuals before inputting it into the evaluation model. This system provides an application programming interface (API) for sending and receiving data with external systems within its external linkage unit. This linkage unit performs mapping processing that automatically converts received external structured data into internal data schemas. The system pipelines these advanced preprocessing and transformation processes. This enables data integration across language barriers, dramatically expanding the learning datasets for machine learning models. Simultaneously, the system maintains the inference accuracy of machine learning while meeting personal information protection requirements. This functionality enables secure and seamless information flow in use cases such as insight-sharing platforms in the global R&D departments of multinational corporations and inter-enterprise collaborative systems that promote open innovation.

[0049] Thus, the embodiments of this invention centrally and multi-targetly utilize generative AI on the server side, automating and highly refining all processes such as natural language processing, knowledge base comparison, data analysis, scoring, reward design, feedback generation, progress reporting, scheduling, meta-analysis, multilingual responses, anonymization, and external collaboration. On the other hand, on the terminal side, it provides employees and managers with a multifunctional user interface that allows for intuitive and real-time input, viewing, and confirmation of information. This significantly improves the fairness, transparency, and efficiency of creative evaluation, maximizing organizational innovation and talent utilization. Specifically, this system integrates multiple specialized neural network models as a microservice architecture. This integrated architecture constructs data pipelines between modules via asynchronous message queues. The system implements terminal-side processing as a single-page application using a virtual document object model. This application minimizes communication with the server, only asynchronously retrieving the necessary differential data to update the screen. Through these architectural designs, this system dramatically improves the scalability and fault tolerance of the entire system. Simultaneously, this system leverages the technical effects of reducing network bandwidth consumption and maximizing throughput based on massively parallel processing. This feature provides stable performance and high availability in use cases such as building large-scale enterprise systems in cloud-native environments or in company-wide activities where traffic spikes are anticipated.

[0050] (System Composition) The system involved in this implementation includes an idea receiving department, an automatic evaluation index calculation department, a business performance integration department, a visualization display department, a compensation design department, a feedback generation department, a progress report department, a database management department, a user authentication department, and an external linkage department.

[0051] The Creative Receiving Department provides an input interface for employees to submit new ideas from terminals such as laptops, tablets, and smartphones. It receives diverse data including the title, detailed description, related materials (PDFs, images, videos, design drawings, reference URLs, etc.), submitter information (employee ID, department, position, contact information, etc.), submission date and time, related project name, envisioned effects and topics, and external information for reference. For example, when employees input "a proposal for an AI-based automated FAQ response system within the company," "an SNS campaign plan for acquiring new customers," or "a delivery route optimization algorithm for reducing logistics costs," they can also attach related drawings, market research data, and case studies of competitors. The receiving department sends this data via encrypted communication and records it in the database management department on the server side. Furthermore, as an input assistance function, the receiving department also enables automatic retrieval and display of similar previously submitted ideas, automatic saving of input content, previewing before submission, editing and appending records after submission, version management of submitted content, and automatic addition of tags upon submission. Specifically, this receiving department has an input processing module that acquires text data in a specified encoding format. This receiving unit performs preprocessing on the additional image data, converting it into a feature map using a convolutional neural network. This receiving unit establishes a secure communication channel using a transport layer security protocol. This receiving unit exchanges session keys using public-key cryptography, encrypts the data, and sends it. As an input aid function, this receiving unit feeds the text stream into a recurrent neural network in real time. This network maintains the hidden state of past input sequences and outputs the probability distribution of the next word's occurrence. This receiving unit provides suggestions using the highest-probability word list as candidates for auto-completion. This system performs efficient initial processing of multimodal data. Therefore, this system significantly reduces the computational load in the subsequent analysis modules. Simultaneously, this system achieves the technical effect of ensuring tamper-proof and confidentiality based on secure data transmission. This function provides a secure and efficient data input environment in use cases such as internal company systems collecting highly confidential unpublished patent information, or new product planning databases requiring strict access control.

[0052] The evaluation metrics are automatically calculated using generative AI on the server side. This analyzes the detailed descriptions and supporting materials of submitted ideas, automatically calculating multiple evaluation metrics such as innovativeness, feasibility, and economic impact. For example, generative AI uses natural language processing to extract the main idea and characteristics of the idea, comparing it with internal and external knowledge bases, patent databases, academic paper databases, industry news, publicly available information from competitors, and past creative cases within the company to determine its differences and novelty from existing technologies. For instance, it automatically determines "which part is novel compared to the existing internal FAQ system," "whether the same technology is already widespread in the industry," and "whether it overlaps with past internal proposals." In the feasibility assessment, it refers to the company's technical resources, past project success rates, failure cases, the company's expert network, external technological trends, and legal and regulatory information to extract the necessary conditions and risk factors for implementation and score them. In the economic impact assessment, it uses market size forecast data, the company's financial performance, industry average cost structure, revenue data from similar past cases, and ROI simulations to automatically calculate economic indicators such as increased sales, cost reduction, brand value enhancement, and improved customer satisfaction. For example, in the case of an "SNS campaign plan," the system refers to past SNS campaign performance data and competitor case studies to predict the expected reach, customer acquisition, and cost-effectiveness. The evaluation metric calculation department automatically weights these metrics and calculates a comprehensive evaluation score. A specific example of a prompt for generative AI to read could be: "Analyze the following creative statement and evaluate it from three perspectives: innovation, feasibility, and economic impact, using a maximum score of 10. Calculate the comprehensive evaluation score along with the supporting evidence for each perspective. Creative statement: 'A proposal for an AI-based automated FAQ response system within the company. For complex inquiries that the existing FAQ system cannot handle, utilize natural language processing technology to improve the accuracy of automated responses.'" Specifically, this calculation department possesses a large-scale language model with hundreds of billions of parameters as its evaluation engine. The department converts the text data containing the prompt into an array of integer identifiers using a tokenizer. The department assigns positional encoding to this array to generate an input tensor. Specific examples of input data could be the indicator vector for evaluating innovation and the word embedding vector of the creative text. This large-scale language model uses a self-attention mechanism to compute long-distance word dependencies within the input text in parallel. The output data from this model consists of continuous scores ranging from 0 to 1, generated from the output nodes corresponding to each evaluation viewpoint via a sigmoid function, and a probability distribution of the tags used to generate the text. Specific examples of the output data include an innovation score of 0.85 and a feasibility score of 0.55. The computational unit multiplies the output score vector by a defined weight matrix to calculate the overall evaluation score.If a score falls below a threshold, this calculation unit performs subsequent processing by raising a flag prompting reconsideration. This system guarantees the objectivity of its evaluations based on mathematical modeling of complex contexts. Therefore, this system enables instantaneous comparison with massive amounts of external data that are beyond human processing, resulting in a significant leap in evaluation accuracy. This function simplifies and enhances the review process in use cases such as the initial screening of investment cases in venture capital or automated review systems for grant applications in public institutions.

[0053] The Business Performance Integration Department manages and integrates diverse performance data related to employees' routine business operations on the server side, combining it with evaluation metrics to calculate overall contribution. For example, it automatically collects various data sources via API linkage or batch processing, including business evaluation scores (regular evaluations by supervisors or colleagues), project achievement (delivery deadline compliance, deliverable quality, customer feedback), attendance data (attendance rate, overtime hours, leave status), customer satisfaction scores, internal and external awards, and skills training records. For instance, "Employee A receives a high evaluation for a new service proposal and also demonstrates high performance in routine business operations, therefore their overall contribution is calculated as high," or "Employee B has a high creativity evaluation score but a low business evaluation score, so their overall contribution becomes moderate." Multiple data points are weighted to calculate the integrated score. The weighting model can be customized for each organization, automatically adjusting based on evaluation policies such as "innovation-oriented" or "business stability-oriented." Furthermore, it learns employees' past contribution tendencies and growth curves, enabling future potential evaluation and automatic extraction of promotion candidates. Specifically, this integration department possesses a scoring engine that utilizes gradient boosting decision tree algorithms or multilayer perceptrons. This scoring engine maps heterogeneous data collected from various systems into a common multidimensional vector space. Specific examples of input data include scalar values ​​representing business evaluation scores and time-series arrays representing attendance status. The scoring engine learns the non-linear interactions of feature quantities, dynamically optimizing the weights of each input feature quantity. The output data from this scoring engine consists of a standardized score representing the overall contribution and an importance vector representing the contribution of each input feature quantity. Specific examples of output data include an overall contribution of 0.88 and a project achievement contribution of 0.45. The integration department then performs subsequent processing, updating and writing the calculated scores to the corresponding records in the employee profile database. This system achieves standardization based on heterogeneous data and the standardization of integrated evaluation indicators. Therefore, this system enables the discovery of hidden correlations based on machine learning models, improving prediction accuracy. This functionality supports data-driven strategic personnel management in use cases such as optimal staffing simulation in large-scale organizations or early detection systems for turnover risk.

[0054] The visualization department displays the evaluation and performance data aggregated on the server side in various formats on the dashboard screen on the terminal side. For example, it provides various visualization modes such as creative evaluation trend charts for each employee, creative submission rankings by department, heatmaps of overall contribution distribution, radar charts for each creative submission by evaluation item, Gantt charts of project progress, and innovation trend analysis charts for the entire organization. For instance, it can extract and display with one click "Departments with the most creative submissions this month," "Creatives that received the highest evaluations in the past six months," and "Employees with rapidly rising evaluation scores." On the terminal side, in addition to employees being able to view their own evaluation status, past submission history, and feedback history, supervisors and managers can gain an overview of their subordinates and the entire organization. Specifically, this visualization department has a rendering engine that processes the multidimensional data aggregated on the server side. This rendering engine uses a manifold learning-based dimensionality reduction algorithm to project high-dimensional data into a low-dimensional space. While preserving the local clustering structure of the data, this rendering engine generates 2D or 3D coordinate data. This visualization department sends the generated coordinate data to the terminal in a lightweight data description language format. This display unit drives the web graphics library on the terminal side. Utilizing hardware acceleration, it renders scatter plots and heatmaps containing tens of thousands of data points at a high speed of 60 frames per second. This system optimally distributes computing resources on the server side and rendering resources on the terminal side. As a result, the system improves overall system responsiveness. Simultaneously, this system promotes an intuitive understanding of complex data structures. This functionality provides an interactive data manipulation experience in use cases such as exploratory data analysis tools for data scientists or real-time performance monitoring dashboards in business meetings.

[0055] The Compensation Design Department uses generative AI on the server side to perform multivariate analysis on factors such as overall contribution, past recognition history, employee motivation, organizational compensation policy, budget status, employee career aspirations, and past incentive responses. This automatically designs optimized compensation and recognition content for each employee. Examples include special recognition, monetary bonuses, stock options, promotion recommendations, new project leader appointments, overseas training opportunities, invitations to company events, access to skills enhancement training courses, company-wide promotions, and company-wide points rewards. These proposals combine various incentives based on organizational culture and employee aspirations. For instance, "For employees who have a stronger career aspiration than monetary rewards in the past, prioritize promotion recommendations and project leader appointments," and "When appointing new project leaders, also consider past experience and project success rates." The rationale and expected effects of the compensation design are automatically generated by the generative AI as explanatory text, displayed to employees and supervisors on the terminal. As a specific example of a prompt text that the generative AI can read, one could say, "Considering employee X's overall contribution, past performance record, motivational tendencies, and the organization's compensation policy, design the optimal compensation and recognition plan. Explain the proposal content along with the justification." Specifically, this design department has an optimization module that uses a reinforcement learning model based on a Markov decision process. The input data to this optimization module includes the employee's overall contribution score, a vector of past compensation performance, and a scalar value representing the budget ceiling. Specific examples of input data could be a contribution score of 0.92 and a budget ceiling of 1 million yen. This reinforcement learning model defines the state space as the employee profile and the action space as a combination of compensation plans. The model defines the compensation function as the employee's performance improvement rate and optimizes the strategy using a Q-learning algorithm. The output data from this reinforcement learning model includes a probability distribution vector representing the optimal combination of compensation plans and textual data of the proposal justification based on the generative model. Specific examples of output data could be a promotion recommendation probability of 0.90 and a special bonus probability of 0.10. This design department selects the plan with the highest expected value from the output probability distribution. This design department utilizes the application programming interface (API) of the HR system to automate the subsequent processing of approval workflows. This system rapidly addresses overall optimization issues under budget constraints. Consequently, it effectively ensures the fairness and transparency of compensation distribution based on data-driven methods. This functionality enables highly effective performance evaluation systems in use cases such as automated incentive calculation systems in results-oriented business organizations or dynamic compensation notification platforms for independent engineers.

[0056] The Feedback Generation Department uses generative AI on the server side to automatically generate evaluation results and improvement suggestions. For example, when an idea is highly innovative but has a low feasibility, it suggests specific action plans such as "To solve technical issues, we recommend collaborating with experts within the company" or "Conducting additional market research can potentially improve feasibility." Conversely, when feasibility is high but economic impact is limited, it generates practical suggestions such as "To maximize cost reduction, explore collaboration with other departments" or "As an additional proposal to existing customers, we recommend pilot implementation." Feedback is distributed to employees and supervisors through multiple channels on the client side, including dashboards, email notifications, and chatbot integration. Furthermore, the feedback content learns from employees' past reactions and improvement experiences, evolving into more personalized content. A specific example of a prompt for the generative AI to read could be, "Based on the following evaluation results, we generate specific improvement suggestions and future action plans for the employee. Evaluation results: Innovation 8 points, Feasibility 5 points, Economic Impact 7 points." Specifically, this generation department possesses a language generation model based on a pre-trained Transformer decoder model. This language generation model is fine-tuned using a high-quality feedback case dataset within the company. The input data to this language generation model consists of a numerical vector of evaluation scores and a matrix of attention weights output from the evaluation model. Specific examples of input data include an innovation score of 0.8 and a feasibility score of 0.5. Based on the input vector and matrix data, the language generation model calculates the probability of occurrence of words relevant to the context and generates an article. The output data from this language generation model consists of natural language text data containing improvement suggestions and a list of Uniform Resource Locators (URLs) for relevant company documents. Specific examples of output data include text on market research for proposal implementation and links to relevant past success stories. The generation department checks the generated text for inappropriate expressions using a rule-based filtering module. The generation department then performs subsequent processing, asynchronously pushing the checked text to the display area on the terminal. This system dramatically reduces the workload of management positions through the automatic generation of high-quality feedback. Simultaneously, the system leverages the technical effect of homogenizing proposal quality by learning from past successful patterns. This feature enables the construction of scalable guidance mechanisms in use cases such as learner-oriented automated grading systems in online education platforms or suggestion generation for quality improvement in customer support departments.

[0057] The progress reporting department uses generative AI on the server side to analyze the evaluation status and progress of creative ideas, as well as their predicted impact on business, automatically generating reports that all relevant personnel can view in real time. For example, it generates detailed reports including progress and impact predictions, such as "The new service proposal is currently in the empirical testing phase and is expected to increase sales in the initial evaluation," or "The business process automation solution requires further discussion due to remaining technical issues." These reports are distributed to relevant personnel through dashboards on the terminal side and email notifications. Furthermore, the generative AI analyzes past evaluation data and performance, automatically calculating the business impact prediction of creative ideas and their contribution to future organizational growth, providing specific numerical predictions, such as "If this creative idea is implemented, it is expected to reduce annual costs by 10%," or "There is a possibility of a 20% increase in new customer acquisition." A specific example of a prompt for the generative AI to read could be, "Based on the following progress status and evaluation history of the creative idea, predict future business impact and generate a progress report for relevant personnel." Specifically, this reporting department has a predictive analysis model that uses an autoregressive time series forecasting model. The input data for this predictive analytics model consists of time-series data on task completion rates and matrix data on resource utilization rates obtained from project management tools. Specific examples of input data include arrays of task completion numbers over the past four weeks and vectors of member utilization rates. This predictive analytics model learns from past time-series patterns and extrapolates future values. The output data from this predictive analytics model consists of continuous values ​​from 0 to 1 representing the probability of goal achievement at a specific future point in time, and a list of identifiers for predicted bottleneck tasks. Specific examples of output data include a release probability of 0.75 in three months and task identifier 402 indicating a high risk of delay. This reporting department combines the prediction results with the generated explanatory text. This reporting department performs subsequent processing by dynamically generating report files in portable document format or interactive Hypertext Markup Language format and storing them in the storage area of ​​relevant personnel. This system enables the early detection of project delay risks through data-driven, objective future predictions. Simultaneously, this system fully automates report production, significantly reducing working hours. This feature enables proactive risk management in use cases such as automating progress management in large-scale system integration projects or generating forecast reports on inventory depletion risks in supply chain management.

[0058] The Database Management Department provides unified management of all creative data, evaluation data, performance data, compensation history, feedback history, progress data, and user information on the server side, undertaking the responsibilities of data integrity, security, backup, and access control. For example, it manages all past creative submissions in chronological order, analyzing each employee's submission tendencies and the organization's overall innovation trends. The User Authentication Department uses employee IDs and passwords, biometric authentication, and two-factor authentication for user authentication when accessing data from terminals, controlling the scope of viewable and editable data based on access permissions. For example, general employees can only view their own submissions, superiors can view their subordinates' data, and managers can access all data. The External Collaboration Department collaborates with external knowledge bases and market databases, industry news APIs, external evaluator interfaces, external compensation services, and external HR systems to ensure system scalability and flexibility. For example, when external experts or customers participate in the evaluation process, it provides an interface for external evaluators, automatically adjusting evaluation criteria and feedback content. Specifically, this Database Management Department adopts a hybrid persistence architecture that combines relational and non-relational databases. This database management unit stores time-series data in a columnar database, improving the execution speed of aggregate queries. This authentication unit implements a stateless authentication authority using an open authorization protocol and Web tokens. This authentication unit hashes and stores biometric authentication data, performing secure lookup processing. This linkage unit uses a remote procedure call protocol for high-speed binary communication with external systems. This linkage unit uses protocol buffers to perform data serialization and deserialization. This system eliminates input / output bottlenecks through appropriate data storage selection. Therefore, this system facilitates horizontal scaling of stateless authentication-based servers. Simultaneously, this system leverages the technology of minimizing network overhead based on binary communication. This functionality provides a robust and scalable foundation for use cases such as consumer-facing web services with millions of users or financial trading systems requiring millisecond-level response times.

[0059] Thus, the system involved in this implementation, through the coordinated operation of various departments, comprehensively and efficiently realizes a series of processes to maximize innovation and talent utilization, including diverse employee creative submissions, advanced automatic evaluation based on generative AI, integration with business performance, visualization, optimal compensation design, specific feedback generation, progress reporting, data management, user authentication, and external collaboration. The prompts read by the generative AI are optimized for each step, such as evaluation, compensation design, feedback, and progress reporting. For example, various prompts are prepared for each purpose, such as "Evaluate the innovativeness of this idea by comparing it with industry standards," "Propose the best incentive based on employee X's contribution," "Provide specific improvement suggestions based on the evaluation results," and "Predict future business impact based on progress data and generate a report." Specifically, this system integrates and controls multiple neural network modules centered on a large language model. This control unit converts the input text data into 512-dimensional word embedding vectors. This control unit inputs these vectors into a Transformer model with a self-attention mechanism. This Transformer model calculates the dependencies between the elements of the input vector. This Transformer model outputs a context-weighted hidden state tensor. This system inputs the hidden state tensor into a subsequent task-specific multilayer perceptron. The multilayer perceptron outputs evaluation scores and classification labels. These outputs are then passed to a threshold determination module. This module triggers subsequent reward design processing only if the score exceeds a predetermined threshold. The system also employs a dynamic template synthesis algorithm in generating the prompt text. This algorithm acquires employee attribute data and past evaluation records as structured data. It converts this structured data into a natural language prompt string. This dynamic prompt is then input into a large language model. The large language model generates context-optimized response text. Regular expressions are used to extract specific numerical values ​​and action plans from this response text. The extracted data is stored in a relational database. Through these processes, the system achieves an objective and high-speed evaluation process independent of human subjectivity. This system eliminates evaluation bias and significantly reduces processing time compared to traditional manual methods. This system can be applied to a variety of use cases, such as the automatic evaluation of business improvement proposals in manufacturing and the generation of feedback for new projects in IT companies.

[0060] (Implementation steps) Step 1: Idea Submission and Reception. Employees use a dedicated input screen on the terminal to input the title, detailed description, relevant materials (PDFs, images, videos, design drawings, reference URLs, etc.), submitter information (employee ID, department, position, contact information, etc.), submission date and time, relevant project name, envisioned effects, and research topic of the new idea. Examples include a variety of ideas such as "A proposal for an AI-based automated FAQ response system within the company," "An SNS campaign plan for acquiring new customers," and "A delivery route optimization algorithm for reducing logistics costs." The data input on the terminal is sent via encrypted communication and recorded in the server-side database. This serves as an input aid function, enabling automatic retrieval and display of similar past ideas, pre-submission preview, post-submission editing and appending of records, version management, and automatic tagging upon submission. Specifically, this receiving module passes the input unstructured text data to the natural language processing pipeline. This pipeline performs morphological analysis on the text to generate a tag sequence. This pipeline removes stop words from the tag sequence. This pipeline uses the TF-IDF algorithm to convert the remaining tags into feature vectors. This receiving module inputs additional image data into a convolutional neural network. This convolutional neural network extracts edge and texture features from the image, outputting a 256-dimensional image feature vector. This receiving module combines the text feature vector and the image feature vector to generate a multimodal feature tensor. This receiving module inputs this multimodal feature tensor into a cosine similarity calculation module. This calculation module calculates the similarity score with the feature tensor group of past ideas in the database. This calculation module extracts past ideas with a similarity score of 0.85 or higher. This receiving module displays the extraction result on the terminal screen in real time. This system uses a support vector machine classifier in its automatic labeling function. This classifier receives the multimodal feature tensor as input. This classifier outputs a probability distribution representing the technical field and business category. This system automatically assigns the category with the highest probability as the label. Through the above series of processes, this system achieves structured input data and high-speed retrieval. This system proactively prevents duplicate proposals and improves database storage efficiency. This system is applied to prior art investigation of patent ideas in research and development departments and duplicate checks of new plans in sales departments.

[0061] Step 2: Automatic Calculation of Evaluation Indicators Based on Generative AI. The server-side automatic evaluation indicator calculation department inputs the detailed description and supplementary materials of the submitted idea into the generative AI, automatically calculating multiple evaluation indicators such as innovativeness, feasibility, and economic impact. The generative AI uses natural language processing technology to extract the idea's purpose and characteristics, comparing it with internal and external knowledge bases, patent databases, academic paper databases, industry news, publicly available information from competitors, and past creative cases within the company to determine its differences and novelty from existing technologies. In the feasibility assessment, the system references the company's technical resources, past project success rates, failure cases, the company's expert network, external technological trends, and legal and regulatory information to extract the necessary conditions and risk factors for implementation and assign them scores. In the economic impact assessment, the system uses market size forecast data, the company's financial performance, industry average cost structure, revenue data from similar past cases, and ROI simulations to automatically calculate economic indicators such as increased sales, cost reduction, brand value enhancement, and improved customer satisfaction. As a specific example of a prompt text for generative AI to read, one could be: "Analyze the following creative statement, evaluate it from three perspectives: innovativeness, feasibility, and economic impact, on a scale of 10, and calculate a comprehensive evaluation score along with the basis for each perspective. Creative statement: 'A proposal for an AI-based automated FAQ response system within the company. For complex inquiries that existing FAQ systems cannot handle, natural language processing technology is utilized to improve the accuracy of automated responses.'" Specifically, this computational unit drives a dedicated evaluation neural network that has been fine-tuned for a large language model. This computational unit encodes the creative statement into a 1024-dimensional input tensor. This computational unit feeds this input tensor into the input layer of the evaluation neural network. This evaluation neural network calculates the semantic distance between keywords within the input tensor using a multi-layered self-attention mechanism. This evaluation neural network calculates the inner product of the embedding vector group of relevant documents obtained from an external database with the input tensor. This evaluation neural network calculates a novelty score based on this inner product value. To evaluate feasibility, this evaluation neural network uses a gradient boosting decision tree model with past project data as input. This decision tree model receives numerical vectors such as the project's budget and duration as input. This decision tree model outputs a continuous value between 0 and 1, representing the probability of success. The computational department executes a Monte Carlo simulation engine to evaluate the economic impact. This engine receives the probability distributions of market growth rate and cost reduction rate as input. The engine performs 10,000 trials and outputs the distribution data of expected returns. The computational department standardizes these output scores to generate a comprehensive evaluation vector. This comprehensive evaluation vector is then passed to the subsequent ranking processing module. Through these high-dimensional operations, this system achieves objective and quantitative evaluations that surpass the limits of human cognition. This system dramatically improves the accuracy of evaluations, contributing to the maximization of investment returns.This system is used for automatic screening of investment cases in venture capital and feasibility studies of new business proposals in the manufacturing industry.

[0062] Step 3: Business Performance Data Integration. The server-side Business Performance Integration Department integrates and manages diverse performance data related to employees' routine business operations, combining it with evaluation indicators to calculate overall contribution. This involves automatically collecting various data sources via API linkage or batch processing, including business evaluation scores (regular evaluations by supervisors or colleagues), project achievement rates (delivery deadline compliance, deliverable quality, customer feedback), attendance data (attendance rate, overtime hours, leave status), customer satisfaction scores, internal and external awards, and skills training records. For example, "Employee A receives a high evaluation for a new service proposal and also demonstrates high performance in routine business operations, therefore their overall contribution is calculated as high," or "Employee B has a high creativity evaluation score, but due to a low business evaluation score, their overall contribution becomes moderate." Multiple data points are weighted to calculate the overall score. Specifically, this integration department standardizes data collected from heterogeneous data sources to construct a single performance feature vector. This integration department scales numerical data such as business evaluation scores and attendance data to a range of 0 to 1. This integration department converts categorized data such as commendation records and training / attendance records into binary vectors using one-hot encoding. These vectors are then combined to generate a 256-dimensional performance feature vector. This performance feature vector is then combined with the creative comprehensive evaluation vector calculated in the previous stage. This combined vector is input into a comprehensive evaluation model composed of a multilayer perceptron. This comprehensive evaluation model has three hidden layers, each applying the ReLU activation function. The model learns the nonlinear interactions of the features within the input vector. The model outputs a comprehensive contribution score ranging from 0 to 100 from the final output layer. Simultaneously, it outputs an attention weight vector representing the influence of each feature on the comprehensive score. This comprehensive contribution score is sent to the subsequent reward design module. The attention weight vector is also sent to the visualization module as a basis for evaluation. Through this series of processes, the system models complex evaluation structures that cannot be fully captured by simple linear addition. This system improves the fairness and transparency of the evaluation, contributing to increased employee motivation. This system is applied to the automatic discovery of the next generation of leaders in human resources departments and the optimal allocation of incentives in sales departments.

[0063] Step 4: Visualization of Evaluation and Contribution. The server-side visualization department displays the aggregated evaluation and performance data in various formats on the dashboard screen on the terminal side. It provides a variety of visualization modes, including creative evaluation trend charts for each employee, creative submission rankings by department, heatmaps of overall contribution distribution, radar charts for each creative submission by evaluation item, Gantt charts of project progress, and innovation trend analysis charts for the entire organization. For example, it can extract and display with one click "Departments with the most creative submissions this month," "Creatives that received the highest evaluations in the past six months," and "Employees with rapidly rising evaluation scores." On the terminal side, in addition to employees being able to view their own evaluation status, past submission history, and feedback history, supervisors and managers can get a comprehensive overview of their subordinates and the entire organization. Specifically, this visualization department retrieves time-series evaluation score arrays and category-based summary tensors from the backend database. This visualization department converts this data into structured data in JSON format that can be interpreted by the front-end rendering library. This visualization department uses principal component analysis algorithms to reduce the high-dimensional performance feature vectors to 2-dimensional coordinate data. This visualization unit uses 2D coordinate data to generate a scatter plot representing employee distribution. It executes the K-Means clustering algorithm, classifying employees into clusters with similar performance tendencies. The unit calculates the center coordinates and variance of each cluster, mapping them as color distributions on a heatmap. An autoregressive moving average model is applied to the time-series data. A predicted curve for future creative submissions is calculated and overlaid on the chart. This visualization unit sends this data to the terminal via HTTPS. The browser on the terminal uses WebGL for hardware acceleration to render these charts at high speed. Through this series of processes, the system transforms massive amounts of evaluation data into intuitively understandable visual information. This system improves the speed of managerial decision-making and supports data-driven organizational operations. The system is applied to company-wide innovation dashboards for management and resource optimization tools for department heads.

[0064] Step 5: Reward and Recognition Design Based on Generative AI. The server-side reward design department uses generative AI to perform multivariate analysis on factors such as overall contribution, past recognition history, employee motivation, organizational reward policy, budget status, employee career aspirations, and past incentive responses. This automatically designs personalized reward and recognition content optimized for each employee. For example, based on organizational culture and employee aspirations, it proposes various incentives such as special recognition, monetary bonuses, stock options, promotion recommendations, new project leadership opportunities, overseas training opportunities, invitations to company events, access to skills enhancement training courses, introductions in company promotions, and company-wide points rewards. A specific example of a prompt for the generative AI to read could be: "Considering employee X's overall contribution, past recognition history, motivation, and organizational reward policy, design the optimal reward and recognition plan. Explain the proposal content along with the supporting evidence." Specifically, this reward design department drives a reward optimization agent equipped with a reinforcement learning algorithm. This agent observes the employee's attribute data and overall contribution score as state vectors from the environment. The agent defines the available combinations of incentives as the action space. This agent receives performance changes following past incentive assignments as reward signals. It updates the state-action value function using a Q-learning algorithm. The agent outputs the combination of incentives that maximizes the expected reward relative to the current state vector. The reward design department inputs this output into a large language model. The large language model receives this combination data along with the employee's career aspirations text as prompts. The large language model outputs structured JSON data containing natural language-based proposal rationale and a specific assignment plan. The reward design department parses this JSON data and compares it with the budget management database. The reward design department determines the final reward plan only if budget constraints are met. Through this series of processes, the system automates personalized incentive design, which is impossible with a uniform reward system. This system maximizes employee engagement and helps reduce turnover. This system is applied in global enterprises for reward optimization for employees with diverse values ​​and for dynamic commission design in business organizations.

[0065] Step 6: Feedback Generation Based on Generative AI. The server-side feedback generation department uses generative AI to automatically generate evaluation results and improvement suggestions. For example, when an idea is highly innovative but has a low feasibility, it prompts specific action plans such as "To solve technical issues, we recommend collaborating with experts within the company" and "Implementing additional market research can potentially improve feasibility." Conversely, when feasibility is high but economic impact is limited, it generates practical suggestions such as "To maximize cost reduction, explore collaboration with other departments" and "As an additional proposal to existing customers, we recommend pilot implementation." Feedback is distributed to employees and supervisors through multiple channels, including dashboards on the terminal side, email notifications, and chatbot integration. A specific example of a prompt for the generative AI to read could be, "Based on the following evaluation results, we generate specific improvement suggestions and future action plans for the employee. Evaluation results: Innovation 8 points, Feasibility 5 points, Economic Impact 7 points." Specifically, this feedback generation department uses a conditional text generation model. This department encodes the array of evaluation scores and the category labels of the idea into a conditional vector. The generation department inputs the conditional vector into a Transformer-based decoder network. This decoder network is pre-trained using a corpus of past high-quality feedback text. The decoder network calculates the probability distribution of word occurrences sequentially based on the conditional vector. The decoder network uses a beam search algorithm to generate the text sequence with the highest likelihood. The generation department extracts a list of specific action plans from this text sequence. This generation department compares this list with the company's talent database. The generation department identifies a list of IDs of company experts with skills related to the action plans. The generation department dynamically embeds the expert's contact information link into the feedback text. The generation department sends this rich text data to a chatbot server via a messaging API. The chatbot server sends push notifications to employees' terminals. Through the above series of processes, this system automatically promotes specific and actionable improvement actions, going beyond simple score prompts. This system accelerates the idea refinement cycle, increasing the final commercialization success rate. This system is applied to the automation of mentoring young researchers in R&D departments and the automatic grading of proposals in planning departments.

[0066] Step 7: Progress Reporting and Impact Prediction Based on Generative AI. The server-side progress reporting department uses generative AI to analyze the evaluation status and progress of ideas, as well as predictions of their impact on business, automatically generating reports that all relevant personnel can view in real time. For example, it generates detailed reports containing information such as "The new service proposal is currently in the empirical testing phase and is expected to increase sales in the initial evaluation" and "The business process automation solution requires further discussion due to remaining technical issues," which are then distributed to relevant personnel via dashboards or email notifications on the terminal side. A specific example of a prompt for the generative AI to read could be, "Based on the following progress status and evaluation history of the idea, predict future business impact and generate a progress report for relevant personnel." Specifically, this progress reporting department drives a Long Short-Term Memory (LSTM) network as a time series forecasting model. This progress reporting department takes past project progress milestone data and resource consumption as input to a time series tensor. The LSM network calculates the project's delay probability and the final cost overrun risk from this time series tensor. The LSM network outputs a predicted score representing the degree of achievement at a specific future point in time. This progress reporting department integrates the predicted score and current status data and inserts them into the prompt template. This prompt is then input into a large language model. The large language model generates report text conforming to a professional business document format. This progress reporting department applies a named entity recognition algorithm to this report text. This progress reporting department highlights risk factors and important milestones within the text. This progress reporting department converts the processed text into HTML format and deploys it to the company's internal portal server. Simultaneously, this progress reporting department automatically generates an alert email for managers when the risk score exceeds a threshold, and sends it via an SMTP server. Through the above series of processes, this system automates project progress management, enabling potential risks to be identified early. This system reduces project manager's management time and improves the overall project success rate of the organization. This system is applied to progress tracking in agile development in system development and to the automatic generation of reports for schedule management in the construction industry.

[0067] (specific use case) For example, when an employee submits a new idea aimed at improving internal business efficiency via a terminal, the idea receiving department receives the employee's input: a proposal titled "A Proposal for an AI-Powered Automated Fund Settlement System," along with detailed information such as "Current fund settlement operations involve a lot of manual work, leading to frequent application errors and approval delays. By implementing AI-based automatic receipt reading, automatic application content checking, and optimized approval processes, the workload can be significantly reduced, and the transparency and accuracy of fund settlement can be improved." This is accompanied by relevant documentation including existing business process diagrams, external AI technology explanations and reports, submitter information, submission date and time, the relevant project name "Financial Business DX," and the projected results of "50% reduction in fund settlement time and 80% reduction in application errors." This information is immediately recorded in the database management department on the server side. Based on the input, the receiving department automatically retrieves similar previously submitted ideas such as "Improvement of Fund Settlement Operations" and "AI-Based Transaction Automation," providing these as reference information to the employee. Specifically, the receiving department segments the input text data character by character and uses an N-gram model to generate an initial feature vector. The receiving unit inputs the initial feature vector into a pre-learned language model, outputting a 768-dimensional dense vector that takes context into account. The receiving unit then uses an optical character recognition engine to extract text data from the uploaded PDF file as related material. This extracted text is also converted into a dense vector. The receiving unit integrates the vectors from the title, detailed description, and related material using average pooling to generate a single proposal performance vector. This proposal performance vector is then input into the HNSW index, which uses an approximate nearest neighbor search algorithm. This index retrieves the top 5 similar idea vectors with the closest Euclidean distance from a database of millions of past ideas, in milliseconds. The receiving unit returns the metadata of the retrieval results to the terminal in JSON format. The terminal application analyzes this JSON data and dynamically renders links to similar proposals in the sidebar on the screen. Through this series of processes, the system accurately captures the semantic structure of the input data, achieving high-speed and high-precision similarity retrieval. This system provides employees with an environment where they can instantly refer to past insights, improving the quality of proposals. This system is used for the automatic classification of improvement requests from citizens in autonomous regions and the retrieval of similar events from customer voices in call centers.

[0068] The evaluation metrics automatic calculation department uses generative AI to analyze submitted creative descriptions and supporting materials. Generative AI extracts keywords such as "AI-based automatic receipt reading," "automatic check of application content," and "optimization of the approval process" through natural language processing technology. It compares these keywords with internal and external knowledge bases, past similar proposals, industry news, patent databases, and academic paper databases to determine differences and innovation compared to existing technologies. For example, if a past internal proposal included "OCR-based automatic receipt input," generative AI emphasizes the novelty of "AI-based automatic check of application content" and "optimization of the approval process," using these as evaluation criteria. In the feasibility assessment, risk factors and implementation conditions are extracted by considering factors such as the availability of AI engineers within the company, the feasibility of integration with existing systems, the success rate of past AI implementation projects, collaboration performance with external suppliers, and legal and personal information protection requirements. In the economic impact assessment, annual working hours for expense settlement operations, the effectiveness of past business improvement measures, cost reduction simulations based on AI implementation, and ROI data from similar cases in other companies are used to calculate the quantitative effects of increased sales and cost reduction. As a specific example of a prompt for generative AI to read, one could be: "Analyze the following creative statement, evaluating it out of 10 points based on three aspects: innovativeness, feasibility, and economic impact. Calculate the overall score along with the supporting evidence for each point. Creative Statement: 'A proposal for an automated funding settlement system utilizing AI. Current funding settlement processes involve a lot of manual work, frequently resulting in application errors and approval delays. By implementing AI-based automatic receipt reading, automatic application content checking, and optimized approval processes, the workload can be significantly reduced, and the transparency and accuracy of funding settlement can be improved.'" Specifically, the computational department inputs the extracted keyword group into a graph neural network. This graph neural network constructs a knowledge graph with keywords as nodes and co-occurrence relationships as edges. The graph neural network executes a message passing algorithm on this knowledge graph, updating the feature representation of each node. The computational department uses this updated feature representation to calculate the cosine similarity with patent classification codes and keyword vectors in external databases and papers. The computational department then identifies nodes with low similarity as elements with high novelty. To assess feasibility, this calculation department retrieves engineer utilization rates and skill sets from the company's resource database as a numerical matrix. This numerical matrix is ​​then used as a constraint to execute a linear programming solver. The solver calculates the optimal solution for resource allocation that maximizes the project's probability of completion and outputs its objective function value as a feasibility score. To assess the economic impact, this calculation department constructs a regression model from past financial data. This regression model takes the hour reduction rate as input and outputs a predicted annual cost reduction. These scores are then integrated to generate a final evaluation vector. This evaluation vector is then sent to the subsequent dashboard update module.This system, through the aforementioned series of processes, transforms qualitative creative descriptions into quantitative evaluation indicators based on rigorous mathematical models. Eliminating human bias, this system dramatically improves the accuracy of investment decisions. It can be applied to automate financing reviews in financial institutions and optimize equipment investment plans in manufacturing industries.

[0069] The Business Performance Integration Department automatically collects data on employees' routine business performance, including business evaluation scores from the finance department, project achievement rates (e.g., progress rate and deliverable quality of finance business DX projects), attendance data (e.g., attendance rate and overtime hours), internal and external awards (e.g., past business improvement awards), and skills enhancement training records (e.g., AI technology training records). This data is then integrated with evaluation indicators to calculate a comprehensive contribution. For example, if an employee has achieved high results in past AI-related projects, the weighting is adjusted to improve the reliability of the feasibility assessment. Specifically, the integration department formats the performance data obtained from the HR system into time-series tensor data. This tensor data is then input into a one-dimensional convolutional neural network. The convolutional neural network extracts the performance variation patterns along the time axis, generating a feature map. This feature map is then input into a fully connected layer, outputting a latent vector representing the employee's skill level. This latent vector is multiplied by the creativity evaluation vector for each element, generating an interaction vector. This integration department inputs the interaction vector into the Softmax function to calculate dynamic weighting coefficients for each evaluation item. Using these weighting coefficients, the integration department calculates a weighted average of the evaluation vectors, deriving the final overall contribution score. If this score exceeds a predetermined threshold, the integration department triggers the process of adding the employee to the high-performing employee list. Through this series of processes, the system achieves a dynamic and adaptive evaluation mechanism based on employees' past performance, rather than static evaluation rules. This system fairly evaluates employees' hidden abilities, promoting optimal talent allocation. This system is applied to skills evaluation of engineers in IT companies and performance management of medical professionals in hospitals.

[0070] The visualization department displays evaluation and performance data on dashboards, such as evaluation trend charts for the "Automated Budget Settlement System Proposal," a ranking of creative submissions within the finance department, heatmaps of overall contribution, radar charts by evaluation item, and Gantt charts of project progress. Employees can view their own evaluation status and feedback history, while supervisors and managers can oversee the department's overall innovation trends and each employee's contribution. Specifically, this visualization department asynchronously retrieves the latest evaluation score array from the backend server via a RESTful API. This array data is stored in a frontend state management library. This visualization department dynamically generates SVG elements using a data-driven Document Object Model (DOM) manipulation library. In the radar chart rendering, this visualization department converts the scores of each evaluation item into polar coordinate vectors. This visualization department calculates the area of ​​the polygon connecting the endpoints of this vector and displays it as a comprehensive balance indicator in the tooltip. In the heatmap rendering, this visualization department inputs the overall contribution score into a color scale function and outputs the corresponding RGB values. This visualization department uses these RGB values ​​to update the background color of each cell in real time. This visualization unit monitors user click events and executes a callback function that displays detailed data in a modal window when a specific data point is selected. Through this series of processes, the system transforms complex multidimensional data into an intuitive graphical interface. This system reduces the user's cognitive load and supports rapid data-driven decision-making. This system is applied in retail for store-by-store sales performance analysis and in logistics for efficient delivery route dashboards.

[0071] The Compensation Design Department uses generative AI to perform multivariate analysis on factors such as overall contribution, past recognition history, employee motivation, organizational compensation policy, budget status, career aspirations, and past incentive responses. This automatically designs optimal compensation and recognition plans that align with employee aspirations and organizational culture, such as "special recognition + monetary bonus," "promotion to head of the Finance & Business DX project," "access to AI technology training," and "introduction in company promotions." A specific example of a prompt for the generative AI could be: "Considering employee X's overall contribution, past recognition history, motivation, and organizational compensation policy, design the optimal compensation and recognition plan. Explain the proposal content along with the supporting evidence." Specifically, the Compensation Design Department combines employee attribute data with their overall contribution score to generate an input feature vector. This input feature vector is then fed into a random forest classifier. The classifier learns from past compensation data and employee satisfaction questionnaire results. Based on the input feature vector, the classifier outputs a probability distribution vector representing the acceptance probability of each compensation category. The Compensation Design Department then selects the top three compensation categories with the highest probabilities from this probability distribution vector. The compensation design department inputs the selected category and budget constraint data into a constraint satisfaction problem solver. This solver calculates the specific compensation amount and privilege combination that maximizes the expected satisfaction value within the budget. The compensation design department embeds this calculation result as structured data into a prompt template in a large language model. The large language model analyzes this prompt and generates personalized compensation proposal text for each employee. The compensation design department automatically sends this text to the approval workflow module of the HR system. Through the above series of processes, this system achieves data-driven, objective compensation design. This system prevents compensation mismatch and maximizes employee motivation within a limited budget. This system is applied to the dynamic optimization of incentive plans in sales departments and compensation algorithms in freelance platforms.

[0072] The Feedback Generation Department uses generative AI to automatically generate evaluation results and improvement suggestions. For example, in cases of high innovation but moderate feasibility, it might suggest specific action plans such as "To supplement internal resources for AI technology, we recommend exploring collaboration with external suppliers" or "Regarding the automatic checking function for application content, we hope to identify technical issues related to API integration with existing systems in advance." A specific example of a prompt for the generative AI to read could be, "Based on the following evaluation results, generate specific improvement suggestions and future action plans for the employee. Evaluation results: Innovation 9 points, Feasibility 6 points, Economic Impact 8 points." Specifically, the Feedback Generation Department receives a vector of evaluation scores as input, driving a hybrid inference engine that combines a rule-based expert system and a neural network. This inference engine compares the score vector with a predefined threshold matrix to identify specific evaluation items as weaknesses. The inference engine retrieves text data of past success stories related to the weak item from a vector database. The inference engine combines the retrieved text data with the current creative description to generate a context vector. The Feedback Generation Department then inputs this context vector into a large language model. The large language model uses an attention mechanism to focus on important keywords within the context. The large language model outputs a natural language text sequence containing solutions to specific technical issues and alternative resource allocation methods. This feedback generation unit uses a natural language processing library to extract verb-object pairs from this text sequence. This pair is then converted into the API format of a task management system and automatically registered as a ToDo list. Through this series of processes, the system transforms abstract evaluation results into concrete action instructions. This system facilitates the transition to the execution phase of ideas and prevents project stagnation. This system is applied to automated grading of student reports in educational institutions and automated code review in software development.

[0073] The Progress Reporting Department uses generative AI to analyze the evaluation status and progress of ideas, as well as their predicted impact on business operations. It generates detailed reports including progress and impact predictions, such as "The proposal for an automated budget settlement system is currently in the requirements definition phase. Initial evaluations anticipate a 50% reduction in budget settlement time and an 80% reduction in application errors," or "Due to remaining technical issues related to system integration with AI, further discussion is needed." These reports are distributed to relevant personnel via dashboards and email notifications. A specific example of a prompt for the generative AI to read could be, "Based on the following progress status and evaluation history of the idea, predict future business impact and generate a progress report for relevant personnel." Specifically, the Progress Reporting Department periodically retrieves numerical arrays representing task completion rates and consumed time from the project management database. This array is then input into a Kalman filter algorithm. This algorithm removes observation noise and estimates the true progress state vector of the project. Using this state vector, the department calculates the probability distribution for predicting future scheduled completion dates and final costs. The department then extracts the mean and variance of this probability distribution and formats it as impact prediction data. This progress reporting department inputs the predicted data and the current list of topics into a large language model. The large language model integrates this quantitative and qualitative data to generate a logically structured report text. This progress reporting department applies a sentiment analysis model to the report text. The department checks whether the tone of the text is overly pessimistic or optimistic and automatically adjusts the tone as needed. The final text is then passed to a PDF generation library to output a formatted document. This document is automatically distributed to relevant personnel's mailing lists. Through the above series of processes, this system ensures project transparency and optimizes information sharing among stakeholders. This system is applied to the automatic creation of materials for management meetings and the automatic generation of IR reports for investors.

[0074] Thus, the system of this invention automates and enhances a series of processes, from the submission of specific business improvement ideas by employees to multi-faceted automatic evaluation based on generative AI, integration with business performance, visualization, optimal compensation design, specific feedback generation, and progress reporting, maximizing organizational innovation and talent utilization. Specifically, this system is installed as multiple independent microservices. The system connects the modules via an asynchronous message queue protocol. The system dynamically scales the computing resources of each module based on the input data flow using container orchestration tools. The system offloads the inference processing of the evaluation and prediction models to a parallel computing cluster using GPUs. Therefore, the system responds with low latency even to simultaneous access from tens of thousands of employees and massive data processing requests. The system caches intermediate data generated in each processing step in a distributed in-memory database. Therefore, the system minimizes the data retrieval overhead for subsequent processing. The system builds a continuous integration and continuous delivery pipeline for machine learning models. The system periodically relearns the model weights using newly accumulated evaluation data, continuously improving inference accuracy. This system integrates these advanced computer technologies to achieve processing speeds and scalability unattainable by traditional manual management methods. It accelerates enterprise digital transformation, becoming a source of competitive advantage. This system is applied in large enterprises as a company-wide hackathon management platform and in cloud services that promote open innovation.

[0075] (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".

[0076] (Implementation methods for carrying out the present invention) In embodiments of this invention, the on-site proposal evaluation and management system in the field of factory robotics uses generative AI as its core, comprehensively utilizing diverse information such as insights from on-site employees and engineers, on-site data, operational performance, organizational talent information, and external technical information to create a robust system structure that maximizes on-site-driven innovation. Specifically, this system employs a distributed processing architecture that links edge computing devices with a cloud server. The system performs primary processing of sensor data in edge devices deployed in each robot unit within the factory. These edge devices filter high-frequency vibration and temperature data in real time, extracting feature vectors. The edge devices then send these feature vectors to the cloud server using a lightweight communication protocol. The control unit on the cloud server side integrates the received feature vectors with text data input from employees. This integrated data is then input into a multimodal neural network. This network uses a cross-attention mechanism to fuse the text embedding vectors with the feature vectors from the sensor data. The network outputs a high-dimensional representation tensor that correlates the physical state of the site with tacit human knowledge. This representation tensor is then fed into subsequent evaluation and prediction modules. This system seamlessly integrates data from cyberspace and physical space through the aforementioned series of processes. It captures minute changes on-site, enabling data-driven, refined evaluation of improvement proposals. This system is applied to the optimization of predictive maintenance systems in automotive manufacturing production lines and the automatic evaluation of yield improvement proposals in semiconductor factories, among other applications.

[0077] First, the Creative Receiving Department enables on-site employees or engineers to submit improvement points or new proposals noticed in daily operations or robot applications from a variety of devices within the factory, including PCs, tablets, smartphones, and dedicated terminals. Examples include: optimizing collaborative robot motion patterns; automating sorting algorithms; introducing AI for anomaly detection; energy-saving operation scheduling; wear prediction models for robotic arms; enhancing safety controls when collaborating with operators; redesigning production line layouts; automating maintenance and inspections; improving inter-robot communication protocols; and creating visual dashboards for on-site data. Proposals addressing all on-site issues and areas for improvement are accepted. The Receiving Department receives a wide variety of data, including: title, detailed description, relevant materials (on-site drawings, operation videos, sensor logs, simulation results, external technical documents, reference paper URLs, patent document numbers, etc.), submitter information (employee ID, production line, responsible process, position, contact information, etc.), submission date and time, relevant project name, expected results (productivity increase rate, cost reduction, defect rate reduction, number of safety accidents reduced, energy consumption reduction, etc.), issues and risks, and external information for reference. For example, when an operator submits a proposal to "optimize the motion patterns of collaborative robots using AI to improve the efficiency and safety of sorting operations" along with sensor logs or on-site video, the receiving unit sends this data to the server via encrypted communication. This also enables version management of submitted content, automatic retrieval and suggestions for similar ideas, pre-submission preview, post-submission editing and appending, automatic tagging of submitted content, time-series management of submission history, automatic multilingual translation of submitted content, anonymous submission functionality, and automatic summarization of submitted content. Specifically, the receiving unit segments the input operation video data frame by frame and inputs it into a 3D convolutional neural network. This network extracts spatiotemporal features and outputs a 512-dimensional vector representing the video's motion patterns. The receiving unit converts the time-series data from the additional sensor logs into frequency domain data using a Fast Fourier Transform algorithm. This frequency data is encoded as a spectrogram image and input into an image recognition model to extract feature vectors of abnormal patterns. The receiving unit applies a pre-learned multilingual language model to the text data. This model maps the input text to a common latent space vector independent of language. This receiving unit uses latent space vectors to drive an automatic translation module, outputting text in the specified language. The receiving unit combines these video feature vectors, sensor feature vectors, and text latent vectors to generate an integrated proposal tensor. This integrated proposal tensor is then input into a clustering algorithm to identify similar past proposal groups. The receiving unit extracts the metadata of these identified proposal groups and displays suggestions on the terminal screen. Through the above series of processes, this system transforms unstructured and multimodal field data into a mathematical expression that can be processed by a computer.This system enables the sharing of global insights across language barriers and eliminates duplicate proposals. It can be applied to platforms for sharing improvement case studies in global factories of multinational corporations or to automatically compile safety countermeasure proposals at construction sites.

[0078] The evaluation index automatic calculation department uses generative AI to analyze the submitted creative descriptions or supplementary materials from multiple perspectives. Generative AI extracts the main points or technical features of the proposals through natural language processing, identifying keywords or logical structures such as "AI-based motion optimization," "sensor data utilization," and "safety enhancement measures." Combining image analysis or time-series data analysis, it automatically extracts changes in motion patterns, signs of anomalies, the relative positions of operators and robots, bottlenecks within the process, robot operating rate trends, abnormal stop frequencies, and interference between operator movement and robot movements from supplementary on-site videos or sensor logs. For example, it performs various analyses such as analyzing changes in robotic arm acceleration from sensor logs to detect wear or anomalies, extracting operator movement and robot motion patterns from on-site videos to identify safety risks or efficiency improvements, quantitatively evaluating the cycle time reduction effect when introducing new algorithms from simulation results, and determining the existence of similar technologies or novelty from external technical documents or patent literature. Generative AI is compared with internal and external knowledge bases, past improvement cases, industry technology trends, patent information, academic papers, on-site application performance data, and external safety specifications. Each viewpoint is scored out of 10 based on its innovativeness (difference or novelty from existing technologies), feasibility (on-site technical resources, implementation costs, schedule, maintenance system, legal compliance, etc.), productivity and quality impact (reduced cycle time, lower defect rate, increased uptime, improved yield, etc.), and safety (reduced risk of collaboration with operators, anomaly detection accuracy, emergency stop function, international safety specification compliance, etc.). The supporting evidence for each viewpoint is automatically generated. For example, prompts may include: "Compared to existing collaborative robot application cases, the motion pattern optimization method is novel, and the safety enhancement strategy that utilizes on-site sensor data is highly original," "The on-site robot control system has high compatibility with existing AI algorithms, and the implementation cost is within acceptable limits," "Simulation confirmed a 15% reduction in the cycle time of sorting operations," and "The risk of interference between operator movement and robot movements is reduced." Furthermore, AI can divide the evaluation process into multiple stages, automatically designing phased evaluation processes such as initial evaluation after submission, re-evaluation after on-site testing, and effect verification evaluation after importation. Specifically, this computational unit inputs frame images extracted from the on-site video into the pose estimation algorithm. This algorithm outputs the operator's skeletal coordinate data and the robot's manipulator joint coordinate data as time-series tensors. This computational unit inputs this coordinate tensor into a graph convolutional network. This network calculates the physical distance or approach speed between the operator and the robot, outputting a safety score representing the collision risk. This computational unit inputs acceleration data from sensor logs into an autoencoder. This autoencoder has pre-learned using normal data and calculates the error between the input data and the reconstructed data.This calculation unit outputs the reconstruction error as an anomaly score, quantifying the signs of wear or failure. It vectorizes the technical keywords extracted from the proposal text and performs cosine similarity retrieval on the vector index of the patent database. The calculation unit calculates the proportion of technical elements with similarity below a specified threshold, outputting it as an innovation score. This calculation unit integrates these safety scores, anomaly scores, and innovation scores and inputs them into a multilayer perceptron. The multilayer perceptron outputs a final 10-level evaluation score for each viewpoint. This calculation unit embeds this evaluation score into prompts in a large language model. The large language model generates a natural language analysis report text as the basis for the score. This calculation unit saves the report text to the database and sends a notification to subsequent approval workflow modules. Through the above series of processes, this system automates high-precision video or waveform analysis that is impossible for the human eye. This system guarantees the objectivity of the evaluation while improving on-site safety and productivity. This system is applied to the optimization of automated handling robots in logistics warehouses or the automatic evaluation of hygiene management processes in food factories.

[0079] The Business Performance Integration Department automatically collects a wide variety of performance data from on-site employees or engineers, including productivity metrics (e.g., number of processes per hour, in-process work efficiency), quality management scores (e.g., defect rate, in-process inspection pass rate), safety compliance rates (e.g., accident-free days, number of alarm reports), past awards and recognitions (e.g., business improvement awards, quality improvement awards, safety awards), skills training and workshop attendance (e.g., AI technology training, robot maintenance workshops, on-site safety training), on-site leadership experience, participation in on-site improvement activities, on-site communication scores, on-site peer evaluations, inter-site transfer history, and on-site multi-skilling. This data is then integrated with AI-based creativity evaluation scores to calculate a comprehensive contribution score. For example, if an operator maintaining high productivity and safety compliance submits an excellent proposal, their comprehensive contribution score increases, and their priority for compensation or recognition rises. The weighted model can be customized according to on-site or organizational policies, such as being automatically adjusted based on evaluation policies like "innovation-oriented," "safety-oriented," "quality-oriented," and "on-site contribution-oriented." Furthermore, AI also learns from employees' past contribution tendencies or growth curves, enabling automatic extraction of future potential evaluations or promotion candidates, and on-site talent allocation optimization proposals. Specifically, this integration department normalizes productivity indicators or quality management scores obtained from the Manufacturing Execution System (MES) to generate multi-dimensional performance vectors. This performance vector is then input into a recurrent neural network. The network learns the time-series changes in performance over the past few years and outputs a gradient vector representing the employee's growth rate. This gradient vector is combined with the current performance vector to calculate a potential ability score. This integration department converts categorical data representing the organization's evaluation policy into one-hot vectors. This one-hot vector is then input into an attention network. The network outputs a dynamic weight coefficient matrix for each element of the performance vector based on the evaluation policy. This integration department integrates this weight coefficient matrix with the performance vector and the vector of creativity evaluation scores through matrix multiplication. This integration department calculates the final comprehensive contribution score from the result of this operation. This comprehensive contribution score and potential ability score are then input into a clustering algorithm. This algorithm categorizes employees into categories such as next-generation leadership candidates or experts in specific technologies. The integration department sends the classification results to the talent allocation optimization module. This module matches the skill requirement matrix of each process with the employee classification results, outputting the optimal personnel allocation plan. Through the above series of processes, this system goes beyond simple past performance evaluation to achieve advanced talent evaluation that includes future growth potential. This system supports talent development that aligns with organizational strategy, maximizing labor productivity. This system can be applied to skills transfer among skilled technicians in factory engineering or dynamic personnel allocation in multi-variety, low-volume production lines.

[0080] The visualization department displays evaluation or performance data in a variety of formats on dashboards. For example, it provides evaluation trend charts for each proposal, rankings of creative submissions by site and by operator, heatmaps of overall contribution, radar charts by evaluation item, Gantt charts of improvement proposal progress, innovation trend analysis charts for the entire site, safety score trends for each site, visualizations of communication networks within sites, creative flow charts between sites, distribution of awards and honors for each site, skill charts for each site, and many other visualization modes. Managers can clearly understand which proposals are most prevalent in which sites, which operators have contributed to innovation, which evaluation items are weaker overall, the flow of insights between sites, and the maturity of the safety culture in each site. Employees can view their own evaluation status or feedback history, past submission history, their position within the site, and information that can be used for self-growth or career development. Specifically, this visualization department has a dedicated rendering engine built on a server. This rendering engine converts multi-dimensional evaluation data vectors obtained from the database into two-dimensional or three-dimensional coordinate data for display. In this transformation process, this system executes dimensionality reduction algorithms such as t-SNE or UMAP. While maintaining the distance relationships between data points in the high-dimensional space, this processing unit maps the data to the low-dimensional space. Using the mapping result, the system generates a heatmap of comprehensive contributions or a node link diagram of the communication network. The visualization unit serializes the generated chart data into a JSON-formatted drawing object. This serialized data is then sent to the client terminal. The browser on the client terminal calls graphics APIs such as WebGL based on the received data. Utilizing GPU hardware acceleration, the terminal renders tens of thousands of data points on the screen without latency. Through this series of processes, the system effectively visualizes large amounts of performance data or evaluation resumes in real time. The processing unit dynamically recalculates the display data based on filtering conditions specified by the manager. By utilizing memory caching, the system reduces the number of database queries associated with recalculation. This configuration significantly reduces server CPU load and network bandwidth consumption. The system demonstrates extremely high responsiveness in use cases where thousands of employees are simultaneously monitored in large-scale manufacturing environments. This visualization department provides personalized views for individual employee dashboards. The system verifies security tokens representing each employee's access permissions, strictly controlling the scope of accessible data. Through this access control, the processing department securely provides necessary feedback while preventing information leakage. This system does not simply mimic human visual cognition; instead, it uses mathematical dimensionality reduction of high-dimensional data and the parallel rendering processing of GPUs—a uniquely computer-based technology—to fundamentally improve the efficiency and speed of information delivery.

[0081] The Compensation Design Department uses generative AI to perform multivariate analysis on factors such as overall contribution, past recognition history, employee or engineer motivation, on-site compensation policies, budget status, career aspirations, past incentive responses, leadership experience on-site, participation in on-site improvement activities, on-site talent allocation, and the distribution of compensation history for each site. This automatically designs diverse compensation and recognition plans corresponding to the aspirations of the site or individual, including special awards, bonuses, promotions to on-site improvement project leaders, participation in joint development with robot manufacturers, skill certification points, participation in skills enhancement training, introductions in company publicity, invitations to on-site events, inter-site transfer recommendations, granting new project initiation authority on-site, and on-site mentorship recommendations. For example, employees with strong career advancement aspirations are prioritized for project leader promotions or skill certification points; employees who value monetary rewards are designed with bonuses or special awards. The AI ​​also automatically generates explanatory text for the basis or expected effects of the compensation design and provides prompts to employees or managers. Specifically, this Compensation Design Department possesses a large-scale language model based on an autoregressive Transformer architecture and a reinforcement learning agent. This system vectorizes employee profile data and inputs it into the AI ​​model. The input data includes a scalar value (e.g., 85) representing overall contribution, a 3D array representing the number of awards received over the past 3 years (a time-series vector representing past awards), and a 768-dimensional vector embedding natural language text representing motivational tendencies. The AI ​​model combines these input vectors and computes the hidden layer representation using a multilayer perceptron. The output data from the AI ​​model includes an array of values ​​representing the output of a Softmax function (e.g., special award 0.1, bonus 0.6, project manager promotion 0.3) representing the probability distribution for each reward plan, and a string representing a text tag column (e.g., "Because of a good response to monetary rewards, a bonus is recommended"). The system utilizes this output in subsequent processing. The processing unit selects the plan with the highest probability distribution value and inserts a record into the reward pre-determining table in the database. The system continuously updates the weight parameters of the AI ​​model using a proximal policy optimization algorithm that uses past reward assignments and subsequent performance changes as reward signals. This system eliminates the subjective and time-consuming compensation decision-making process performed by human managers, replacing it with objective and instantaneous optimization calculations based on multidimensional data. Through this process, the system achieves efficient allocation of computing resources and leverages the technological advantage of drastically reducing processing time through the automation of personalized compensation design. In use cases where individual compensation plans are generated in real time for thousands of employees, this system fundamentally improves the data processing capabilities of human resource management systems.

[0082] Based on AI-based evaluation results, the Feedback Generation Department automatically generates specific improvement suggestions or action plans, such as "The likelihood of implementing the action pattern optimization solution is increased by analyzing the sensor data on site in more detail," "To strengthen the safety evaluation, it is recommended to implement additional risk assessments," "Additional learning data collection is effective in improving the accuracy of anomaly detection algorithms," "Energy-saving operation proposals maximize their effectiveness by linking with on-site operation scheduling," "Installation costs can be suppressed by promoting collaboration with other departments on site," and "Improved safety can be expected by linking with on-site operator training programs." These suggestions are then distributed to operators or engineers. Feedback is distributed through multiple channels, including dashboards, email notifications, chatbot integration, on-site bulletin board integration, and automatic generation of on-site morning meeting materials. It also learns from employees' past reactions and improvement experiences, evolving into more personalized content. Specifically, this Feedback Generation Department possesses a large-scale language model dedicated to natural language generation. The system embeds the score data output from the evaluation module into prompt word templates and inputs them into this AI model. The input data to this AI model includes a 4-dimensional vector representing evaluation results (e.g., innovativeness 9, feasibility 6, productivity 8, security 7), a 512-dimensional numerical array representing the embedding vector of the proposal text, and a vector of past feedback history. This AI model uses an attention mechanism to weight important features within the input data, calculating the probability distribution of the next word while maintaining context. The output data from this AI model includes a text tag column representing improvement suggestions (e.g., "Improving feasibility by increasing sensor sampling rate"), and a numerical array representing the probability vector for selecting distribution channels (e.g., email 0.8, dashboard 0.2). This output is utilized in subsequent processing. The processing unit converts the generated text into the API format of each distribution channel and sends it to a message queue. The system uses an asynchronous task worker to retrieve data from the message queue and sends HTTP requests to external mail servers or chat systems. This configuration prevents system blocking caused by synchronous communication and improves throughput for a large number of feedback generation requests. The system continuously improves the generation quality of this AI model by fine-tuning the cross-entropy error function. This system leverages the reduced latency and increased personalized accuracy resulting from optimizations in natural language generation processing. In use cases where specific improvement actions are immediately suggested to on-site operators, this system eliminates communication delays and dramatically improves business efficiency.

[0083] The progress reporting department uses AI to analyze the evaluation status of proposals or the progress and business impact predictions of on-site implementation. Quantitative progress and effect predictions, such as "The motion optimization proposal is currently in the on-site testing phase, and the initial evaluation expects a 15% increase in sorting efficiency and a 20% reduction in the accident rate," "The anomaly detection AI implementation plan requires further discussion due to technical issues related to the linkage with the residual system," "The energy-saving operation proposal is expected to reduce monthly energy consumption by 10%," "Simultaneous implementation on multiple production lines on-site is underway, and the effect comparison between production lines is automatically summarized," and "Safety improvement measures linked to on-site operator training programs are underway," are distributed as reports to relevant personnel. Progress reports are distributed to a variety of relevant personnel, including on-site managers, management, relevant departments, and external partners, with automatically adjusted viewing permissions or report granularity. Specifically, this progress reporting department possesses a Long Short-Term Memory network or a Transformer-based time series forecasting model. The system converts time series progress data extracted from the database into tensor format and inputs it into this AI model. The input data to this AI model includes time-series tensors representing state changes over the past 30 days as past progress logs, integer values ​​representing the current state phase as indicators, and floating-point numbers representing an initial business impact, such as a 15% efficiency improvement rate as initial predictions. This AI model learns long-term dependencies within the time-series data and regresses to predict future values. The output data from this AI model includes a vector of continuous values ​​representing future business impact predictions, such as an 18% efficiency improvement rate three months later, and a string such as "Progressing smoothly, further efficiency improvements expected" as report text. This output is utilized in subsequent processing. If the predicted value falls below a specified threshold, the processing unit executes a branch process that issues an alert to managers. The system performs filtering based on a viewing permission matrix, dynamically masking confidential information in the report according to the user's position or department. This configuration leverages the increased processing speed through parallelization of predictive computation and the enhanced security through automated access control. This system completely eliminates manual data aggregation and report creation tasks, generating reports instantaneously using a data-to-text conversion algorithm. In use cases where management can monitor the return on investment of projects in real time and make rapid decisions, this system ensures zero delay in information transmission.

[0084] Furthermore, for example, to cope with multilingual environments on-site, generative AI can automatically translate submitted ideas or feedback, and can also smoothly share and evaluate ideas across global sites. For example, to ensure employee anonymity, generative AI can be installed to mask submitter information and eliminate bias in evaluation. For example, when external robot manufacturers or experts participate in the evaluation process, generative AI can automatically adjust evaluation benchmarks or feedback content, providing an interface for external evaluators. For example, when multiple ideas are being developed simultaneously on-site, mechanisms can be installed to visualize the progress of each evaluation or feedback history and time-series status, providing an overview of the entire site's innovation activities. For example, to foster a safety culture on-site, AI can be installed to analyze the frequency or content of safety-related proposals, automatically evaluate the safety culture maturity of each site, and promote the sharing of best practices among sites. Specifically, this system has a multilingual translation model based on an encoder-decoder Transformer architecture and a named entity extraction model. The system tokenizes the input text data and inputs it into this AI model. The input data to this AI model includes an array of character codes as multilingual text, such as a proposal text representing Japanese, and integer values ​​as metadata, such as the source language ID or target language ID. The AI ​​model generates a context vector for the input language by an encoder and an autoregressive column of tags for the target language by a decoder. The output data from this AI model includes an array of character codes as translated text, such as converted to English, and a masked text, such as strings where personal or department names are replaced with asterisks. This output is used in subsequent processing. The processing unit saves the translation results to an in-memory cache server, omitting re-translation of the same text. The system stores the masked data in a dedicated anonymized database, providing only this anonymized data to access requests from the evaluation module. This configuration leverages the technical effects of reduced computational resources and data privacy guarantees resulting from the integration of natural language processing pipelines. In the global manufacturing industry, this system significantly improves the versatility and security of the system in use cases such as real-time sharing of technical proposals across language barriers and implementing impartial reviews free from evaluation bias.

[0085] Thus, the embodiments of this invention integrate the insights of on-site employees or engineers with on-site data, consistently supporting all on-site innovation activities based on generative AI, including multi-faceted automatic evaluation, integration, reward design, feedback, progress management, multilingual responses, anonymization, external collaboration, inter-site insight exchange, and the cultivation of a safety culture. This results in a highly practical and scalable system capable of efficiently and fairly promoting on-site-led, continuous improvements in productivity, quality, and safety in the field of factory robotics. Specifically, this system adopts a microservice architecture, deploying each functional module as an independent container. The system uses container orchestration tools to dynamically scale the number of replicas of each module based on traffic increases or decreases. The system employs an asynchronous publish-subscribe model using message queues for inter-module communication. This configuration achieves fault tolerance, preventing the entire system from stopping even if a specific module fails. The system accumulates on-site sensor data, employee input data, and data obtained from external APIs into an integrated data lake, continuously feeding it into the relearning pipeline of each AI model. Through the automation of this data flow, the system builds a self-learning mechanism that improves the inference accuracy of AI models over time. This system replaces evaluation and management, which are traditionally performed by human mental processes, with purely computer-based processes such as matrix operations in high-dimensional vector spaces, probability distribution calculations, and asynchronous message passing. Through this replacement, the system achieves clear technical improvements, including increased processing speed, enhanced data management, increased computational efficiency, and reduced communication load. In large-scale smart factories, in complex environments where tens of thousands of robots and thousands of employees collaborate, this system demonstrates exceptional practicality and scalability in innovative, cyclical automation and continuous optimization at the system level.

[0086] (System Composition) The system described in this embodiment includes an idea receiving unit, an automatic evaluation index calculation unit, a business performance integration unit, a visualization display unit, a compensation design unit, a feedback generation unit, a progress reporting unit, a database management unit, a user authentication unit, and an external linkage unit. Specifically, this system runs on a distributed computing cluster built within a virtual private cloud environment. To execute the software modules constituting each unit, the system utilizes a physical server cluster equipped with multiple multi-core CPUs, large-capacity RAM, and GPUs dedicated to parallel computing. The system receives external requests via a load balancer and distributes traffic to appropriate server nodes. The database management unit employs a hybrid configuration of relational databases, which excel in transaction processing, and NoSQL databases, which excel in high-speed retrieval of unstructured data. In the user authentication unit, standard protocols such as OAuth or OpenID Connect are installed, and JSON Web Tokens are issued for stateless session management. In the external linkage unit, an API gateway is configured to perform rate limiting control or payload format conversion. This configuration leverages the technical effects of optimal allocation of hardware resources, establishment of secure communication paths, and reduced latency in inter-system linkage. This system enables future feature additions or AI model updates without downtime through loosely coupled interoperability among its components. Within a globally deployed manufacturing network, it seamlessly integrates with systems at various locations and is optimized for enterprise-level use cases involving real-time processing of massive amounts of data.

[0087] The Creative Reception Department enables on-site employees or engineers to submit improvement points or new proposals noticed in daily operations or robot applications from a variety of devices within the factory, such as PCs, tablets, smartphones, and dedicated terminals. For example, it receives proposals on all on-site issues or areas for improvement, including optimizing the motion patterns of collaborative robots, automating picking operations, introducing AI for anomaly detection, energy-saving operation scheduling, wear prediction models for robotic arms, enhancing safety controls when collaborating with operators, redesigning production line layouts, automating maintenance and inspections, improving inter-robot communication protocols, and creating visualization dashboards for on-site data. The Creative Receiving Department receives a wide variety of data, including titles, detailed descriptions, related materials (on-site drawings, operation videos, sensor logs, simulation results, external technical data, reference paper URLs, patent document numbers, etc.), submitter information (employee ID, production line, responsible process, position, contact information, etc.), submission date and time, related project name, expected results (productivity increase rate, cost reduction, defect rate reduction, number of safety accidents reduced, energy consumption reduction, etc.), research topics and risks, and external information used for reference. It also implements version management of submitted content, automatic retrieval and suggestions for similar creative ideas, pre-submission preview, post-submission editing and appending records, automatic tagging of submitted content, time-series management of submission history, automatic multilingual translation of submitted content, anonymous submission functionality, and automatic summarization of submitted content. Specifically, this Creative Receiving Department has a dedicated API endpoint for receiving multi-part form data from client terminals. The system categorizes the received data, routing text data to a natural language processing pipeline, image data to a computer vision pipeline, and video data to frame extraction and processing. This system utilizes an embedding model that tokenizes text and maps it to a high-dimensional vector space for similar creative retrieval. Input data to this AI model includes, for example, strings representing proposal content as text, 3-channel RGB tensors (e.g., 224 pixels square) as images, and arrays of 30 image tensors as videos. Output data from this AI model includes parsed results in a standardized data structure, such as JSON format, and a similarity score, such as a floating-point number between 0 and 1 representing the cosine similarity to past proposals. This output is used in subsequent processing. If the similarity score exceeds a predetermined threshold, the processing unit performs a branching process, displaying a warning screen for duplicate proposals to the user. This system uses an approximate nearest neighbor search algorithm to retrieve similar proposals from millions of past data points in milliseconds. This configuration leverages the technical advantages of reducing the load on subsequent processing through input data standardization and saving storage capacity by eliminating duplicate proposals. On-site operators can directly upload videos taken with their smartphones; the system automatically extracts features and registers them in the database, ensuring data quality while eliminating the hassle of input.

[0088] The evaluation index automatic calculation department uses generative AI to analyze the submitted creative descriptions or supplementary materials from multiple perspectives. Through natural language processing, it extracts the main points or technical features of the proposals, specifically targeting keywords or logical structures such as AI-based motion optimization, sensor data utilization, and safety enhancement measures. Combining image analysis or time-series data analysis, it automatically extracts changes in motion patterns, signs of anomalies, the relative positions of operators and robots, bottlenecks within processes, robot operating rate trends, frequency of abnormal stops, and interference between operator movement and robot movements from supplementary on-site videos or sensor logs. Furthermore, it compares these findings with internal and external knowledge bases, past improvement cases, industry technology trends, patent information, academic papers, on-site application performance data, and external safety specifications. Each aspect—innovation, feasibility, impact on productivity and quality, and safety—is scored out of 10, automatically generating the basis for each assessment. For example, various analyses can be performed, such as analyzing changes in the robot arm's acceleration from sensor logs to detect wear or abnormal signs, extracting operator movement patterns and robot motion patterns from on-site videos to identify specific safety risks or efficiency improvements, quantitatively evaluating the cycle time reduction effect when introducing new algorithms from simulation results, and determining the existence of similar technologies or novelty from external technical data or patent literature. As a specific example of a prompt for generative AI to read, one could say, "Analyze the following robot application improvement proposal, evaluating it on a scale of 10 out of 10 from four perspectives: innovativeness, feasibility, productivity and quality impact, and safety. Calculate the overall evaluation score along with the basis for each perspective. Proposal description: 'Improve the efficiency and safety of picking operations by optimizing the robot's motion patterns in collaboration with operators through AI. Sensor logs and on-site videos are attached.'" Specifically, the automatic calculation unit of this evaluation index possesses a multimodal Transformer model with a cross-attention mechanism for text and images. This system converts various pre-processed data into tensor form and inputs it into this AI model. The input data to this AI model includes, for example, an array of integer IDs segmented from the proposal text as text marker columns, an image feature vector such as a 2048-dimensional vector extracted from a convolutional neural network, and a 1-dimensional array of 3-axis accelerations sampled at 100 Hz as time-series sensor data. This AI model projects these heterogeneous data into a common latent space and calculates the correlations between the modalities. The output data from this AI model includes, for example, floating-point numbers representing innovation (8.5) as scores for each evaluation point, and a string such as "because a novel algorithm was used" as supporting text. This output is utilized in subsequent processing. The processing unit calculates a weighted average of the scores to obtain a comprehensive evaluation score and updates the evaluation table in the database.This system employs supervised learning, using past expert evaluation data as teacher data, and optimizes the weight parameters of the AI ​​model using a mean squared error function. This architecture leverages the technical advantages of improved evaluation accuracy through the integrated analysis of multimodal data and significantly reduced processing time by replacing manual human evaluation. In use cases involving complex robot control algorithm proposals, this system eliminates bottlenecks in the review process by instantly analyzing accompanying simulation logs and objectively scoring the technical appropriateness.

[0089] The Business Performance Integration Department automatically collects a wide variety of performance data from on-site employees or engineers, including productivity metrics (processing per hour, in-process efficiency), quality management scores (defect rate, in-process inspection pass rate), safety compliance rates (accident-free days, number of alarm reports), past awards (business improvement awards, quality improvement awards, safety awards), skills training and workshop attendance (AI technology training, robot maintenance workshops, on-site safety training), on-site leadership experience, participation in on-site improvement activities, on-site communication scores, peer evaluations, inter-site transfer history, and on-site multi-skilling. This data is then integrated with AI-based creativity evaluation scores to calculate a comprehensive contribution score. For example, if an operator maintaining high productivity and safety compliance makes an excellent proposal, their comprehensive contribution score increases, and their priority for compensation or awards rises. The weighted model can be customized according to on-site or organizational policies, automatically adjusting based on evaluation policies such as innovation-oriented, safety-oriented, quality-oriented, and on-site contribution-oriented. AI also learns from employees' past contribution tendencies and growth curves, enabling it to automatically evaluate future potential, extract promotion candidates, and propose on-site talent allocation optimization suggestions. Specifically, the Business Performance Integration Department possesses predictive models based on gradient-boosting decision trees or multilayer perceptrons. The system performs missing value imputation and standardization on various indicators collected from the database, generating feature vectors. The input data to this AI model includes a 50-dimensional numerical vector, such as scores for productivity, quality, and safety, which are linked to various performance indicators. The AI ​​model learns non-linear decision boundaries and scores based on the interactions of each feature. The output data from this AI model includes a continuous value (e.g., 78.5) representing the overall contribution score (ranging from 0 to 100) and a floating-point number (e.g., 0.85) representing the probability of promotion, which is used in subsequent processing. The processing department generates an employee ranking list based on the calculated scores and transmits the data to the compensation design module. The system dynamically recalculates the weighting parameters for each feature through matrix operations based on changes in evaluation policies. This system leverages the technical advantages of standardizing and integrating heterogeneous data to stabilize calculations and improving the objectivity and accuracy of talent evaluation by importing predictive models. In large-scale factories with thousands of employees, this system dramatically improves the utilization of personnel data by automatically proposing optimal talent allocation use cases through real-time integration of individual daily performance and contributions to innovation.

[0090] The visualization department displays evaluation and performance data in a variety of formats on dashboards, providing evaluation trend charts for each proposal, rankings of creative submissions by site and by operator, heatmaps of overall contribution, radar charts by evaluation item, Gantt charts of improvement proposal progress, innovation trend analysis charts for the entire site, safety score trends for each site, visualization of communication networks within the site, creative flow charts between sites, distribution of awards and honors for each site, skill charts for each site, and many other visualization modes. Managers can clearly understand which proposals are most prevalent in which site, which operators have contributed to innovation, which evaluation items are weaker in the organization, the knowledge flow between sites, and the maturity of the safety culture in each site. Employees can view their own evaluation status and feedback history, past submission history, their position within the site, and information that can be used for self-growth and career development. Specifically, this visualization department consists of a front-end application and a back-end streaming data server. The system detects update events in the database and uses the WebSocket protocol to push and distribute differential data to client terminals. This system utilizes a virtual DOM for efficient rendering on the client side, minimizing screen redraw costs. The processing unit uses principal component analysis to convert high-dimensional skill graph data into two-dimensional coordinates and depict it as a scatter plot. In network graph depiction, the system employs a layout algorithm based on a mechanics model to avoid node overlap. The visualization unit adopts a responsive design, dynamically calculating the optimal layout based on device resolution, from large PC monitors to small smartphone screens. This configuration effectively reflects real-time data updates on the screen while minimizing network bandwidth consumption. The system provides an interactive drill-down display function, asynchronously retrieving detailed data when an administrator clicks on a specific chart element on the dashboard. In the use case where a factory manager projects data onto a large monitor in a control room overlooking the innovation status of all network points, the system visually organizes massive amounts of information, enabling an intuitive understanding of the situation.

[0091] The Compensation Design Department uses generative AI to perform multivariate analysis on factors such as overall contribution, past recognition history, employee or engineer motivation, on-site compensation policy, budget status, career aspirations, past incentive responses, leadership experience on-site, participation in on-site improvement activities, on-site talent allocation, and the distribution of compensation history for each site. This automatically designs a variety of compensation and recognition plans corresponding to the site's or individual's aspirations, including special awards, bonuses, promotions to on-site improvement project leaders, participation in joint development with robot manufacturers, skill certification points, participation in skills enhancement training, introductions in company publicity, invitations to on-site events, recommendations for inter-site transfers, granting authority to initiate new projects on-site, and recommendations for mentorship roles on-site. Specific examples of prompts for the generative AI to read include: "Considering employee X's overall contribution, past recognition history, motivation, and on-site compensation policy, design the optimal compensation and recognition plan. Explain the proposal content along with the supporting evidence." Specifically, this Compensation Design Department uses dynamically constructed contextual inputs from prompting engineering techniques into a large-scale language model. This system serializes structured data retrieved from the database into a JSON string and inserts it into placeholders in the prompt template. Input data to this AI model includes, for example, a scalar value of 92 representing overall contribution, an array of past award IDs representing past awards, and a vector representing motivational tendencies, such as numerical representations of questionnaire results. This AI model uses a self-attention mechanism to calculate the dependencies between elements within the input sequence and infer the optimal reward plan. Output data from this AI model includes, for example, an array of values ​​representing the probability distribution for each reward plan (e.g., special award 0.7, bonus 0.2, further education participation 0.1), and a string representing the generated descriptive text (e.g., "Evaluate high contribution and leadership and propose special award"). This system utilizes this output in subsequent processing. The processing unit selects the plan with the highest probability distribution value and sends the data to the approval workflow system. This system uses regular expressions to verify that the output text conforms to the specified format; in case of parsing errors, it adjusts temperature parameters and performs regeneration processing. This system leverages the technical advantage of rule-based programs that do not require writing complex conditional branches to explore optimal solutions that satisfy a wide variety of constraints. In use cases where the system automatically generates reward combinations that maximize employee motivation within a limited budget, it drastically reduces the workload of HR departments.

[0092] Based on AI-generated evaluations, the Feedback Generation Department automatically generates action pattern optimization solutions. This includes: improving feasibility through more detailed analysis of on-site sensor data; recommending additional risk assessments to enhance safety evaluations; effectively improving the accuracy of anomaly detection algorithms through additional learning data collection; maximizing the effectiveness of energy-saving operation proposals by linking them with on-site operation scheduling; reducing installation costs by promoting collaboration with other departments on-site; and expecting safety improvements through linkage with on-site operator training programs. These specific improvement suggestions or action plans are then distributed to operators or engineers. Feedback is distributed through multiple channels, including dashboards, email notifications, chatbot integration, on-site bulletin board integration, and automatic generation of on-site morning meeting materials. It also learns from employees' past reactions and improvement experiences, evolving into more personalized content. A specific example of a prompt that the generative AI can read could be: "Based on the following evaluation results, specific improvement suggestions and future action plans are generated for the employee. Evaluation results: Innovation 9 points, Feasibility 6 points, Productivity and Quality Impact 8 points, Safety 7 points." Specifically, this feedback generation unit uses a few-shot prompting method, incorporating past excellent feedback examples into the context and inputting them into a large language model. The input data to this AI model includes, for example, an array containing scores for each item as an evaluation result vector, coordinate data in a high-dimensional space as an embedding vector for the proposal text, and a vector of past feedback history. The AI ​​model focuses on items with lower input scores, retrieving specific actions from the knowledge base to improve those items and generating text. The output data from this AI model includes a text tag column for improvement suggestions, such as a string like "recommend extending the data collection period," and a probability vector for the selection of distribution channels. This output is utilized in subsequent processing. The processing unit converts the generated text into the API format of each distribution channel and sends it as an asynchronous task to a distributed task queue. The system internally evaluates the quality of the generated text using BLEU or ROUGE scores, regenerating it if the quality benchmark is not met. This configuration achieves the technical effect of eliminating the time required for manual drafting of feedback text by the system, autonomously generating high-quality text. This system functions extremely effectively in use cases where, even if a proposal is rejected, it automatically prompts the reasons and provides specific steps for future improvement to prevent a decline in employee motivation.

[0093] The progress report department uses AI to analyze the evaluation status of proposals or the progress and business impact predictions of on-site implementation. For example, the action optimization proposal is currently in the on-site testing phase, with initial evaluations expecting a 15% increase in picking efficiency, a 20% reduction in the accident rate, and further discussion on the AI ​​implementation plan for anomaly detection due to technical challenges related to the residual system linkage. The energy-saving operation proposal is expected to reduce monthly energy consumption by 10%. Simultaneous implementation on multiple production lines is underway, and inter-production line performance comparisons are automatically summarized. Safety improvement measures linked to on-site operator training programs are also underway. These quantitative progress and effect predictions are distributed as reports to relevant personnel. The progress report is automatically distributed to various stakeholders, including on-site managers, management, relevant departments, and external partners, with adjusted viewing permissions and report granularity. A specific example of a prompt for the generative AI to read could be, "Based on the progress status and evaluation history of the following ideas, predict future business impact and generate a progress report for relevant personnel." Specifically, this progress report department possesses a neural generative model that performs data-to-text conversion algorithms. This system converts structured data retrieved from the database into a flat list of key-value pairs and inputs it into the AI ​​model. The input data includes, for example, an array of time-series status IDs representing past progress logs, current status flags, and initial forecasts of business impact. The AI ​​model analyzes the trends in the input numerical data and generates fluent report text following the grammatical rules of natural language. The output data from the AI ​​model includes, for example, numerical values ​​representing cost reductions as forecasts of future business impact, and strings such as "targets expected to be achieved next month" as report text. The system utilizes this output in subsequent processing. The processing unit renders the generated report from HTML to PDF format and automatically sends it to the email addresses of relevant personnel using the SMTP protocol. This system generates expressive text while ensuring the accuracy of numerical values ​​by combining template-based generation techniques and neural generation techniques. This configuration fully automates the standardized report creation process and achieves real-time information sharing using computer resources. In large-scale manufacturing environments with multiple projects running concurrently, this system integrates the status of each project and automatically generates daily summary reports for management, significantly reducing management costs.

[0094] The Database Management Department centrally manages all creative data, evaluation data, performance data, compensation history, feedback history, progress data, user information, on-site information, and external collaboration data. It is responsible for data integration, security, backup, access control management, data mining, history management, and providing data collaboration APIs. The User Authentication Department authenticates users from terminals using employee IDs or passwords, biometric authentication, or two-factor authentication, controlling the scope of viewable and editable data based on access permissions. The External Collaboration Department collaborates with external knowledge bases or market databases, industry news APIs, external reviewer interfaces, external compensation services, external HR systems, technical information APIs from robot manufacturers, and external security specification databases to ensure system scalability and flexibility. Specifically, the Database Management Department constructs a master-slave replication cluster, distributing read queries to slave nodes to reduce database load. Frequently accessed data is cached in-memory data stores such as Redis, eliminating disk I / O bottlenecks. The User Authentication Department hashes and stores passwords using the Bcrypt algorithm and encrypts the entire communication path using the TLS 1.3 protocol. This system issues an expired JSON Web Token upon successful authentication, which is included in the header of subsequent API requests to achieve stateless authentication. The external communication department has a RESTful API gateway that controls request rate limiting or IP address-based filtering from external systems. This system parses XML data obtained from external APIs into JSON format for internal processing, absorbing inconsistencies in data structure. This configuration achieves the technical benefits of improved data retrieval speed, prevention of unauthorized access, and reduced latency in inter-system communication. In use cases where legacy HR systems within a company can securely and rapidly integrate with the latest cloud-based AI evaluation systems, this system breaks down data silos and provides an integrated data foundation.

[0095] Thus, the system described in this embodiment, through the coordinated operation of its various components, comprehensively and efficiently realizes a series of processes aimed at maximizing field-led innovation and talent utilization in the field of factory robotics, including diverse idea submissions from on-site employees or engineers, advanced automated evaluation by generative AI, integration with business performance, visualization, optimal compensation design, specific feedback generation, progress reporting, data management, user authentication, and external collaboration. Specifically, the components of this system act as independent processes, interacting with each other based on an event-driven architecture. The system publishes the event of an idea submission to a message broker, and the evaluation or notification module subscribes to the event and begins processing asynchronously. This configuration ensures that the scalability and responsiveness of the entire system are not compromised even when specific processing takes time. This system completely reconstructs existing business processes that rely on human subjectivity or manual operations through a combination of computer technologies: vector operations on high-dimensional data, pattern recognition based on neural networks, and parallel processing based on distributed systems. Through this reconstruction, the system achieves significant technical improvements, including faster processing, guaranteed objectivity in evaluation, and real-time processing of massive amounts of data. This system is not merely a tool for improving business efficiency, but rather a sophisticated information processing device that functions as a tightly integrated hardware and software system, encompassing data input, inference, output, and feedback loops. As a core foundation for digital transformation in the manufacturing industry, this system fundamentally strengthens a company's competitiveness by collecting and analyzing all on-site data and autonomously and continuously operating an improvement cycle.

[0096] (Implementation steps) Step 1: On-site Proposal Submission and Reception. On-site employees or engineers can submit proposals from various devices within the factory, such as PCs, tablets, smartphones, and dedicated terminals, covering all on-site issues and potential areas for improvement. These proposals include optimization of collaborative robot motion patterns, automation algorithms for sorting operations, importation of AI for anomaly detection, energy-saving operation scheduling, wear prediction models for robotic arms, enhanced safety controls when collaborating with operators, redesign of production line layouts, automation of maintenance and inspection, improvement of inter-robot communication protocols, and visualization dashboards for on-site data. The submission should include a title, detailed description, relevant materials (on-site drawings, operation videos, sensor logs, simulation results, external technical data, URLs of referenced papers, patent document numbers, etc.), submitter information, submission date and time, relevant project name, expected results, issues and risks, and external information for reference. The receiving department will send this data to the server via encrypted communication. The system also includes version management of submitted content, automatic retrieval and suggestions for similar ideas, pre-submission preview, post-submission editing and appending records, automatic tagging of submitted content, time-series management of submission history, automatic multilingual translation of submitted content, anonymous submission functionality, and automatic summarization of submitted content. Specifically, in this step, the application on the client terminal temporarily stores the user's input data in memory and performs input field validation. If the terminal detects missing required input items or incorrect data format, it immediately renders an error message. The terminal serializes the validated data into a JSON-formatted payload and sends it to the server as an HTTPS request encrypted using the TLS 1.3 protocol. Upon receiving the request, the server-side receiving unit deserializes the payload and performs sanitization processing to prevent attacks such as cross-site scripting. The system also segments large video files or log data into blocks and asynchronously uploads them to distributed object storage. After uploading, the system associates a unique URI issued with the proposal record in the database and saves it. This configuration effectively utilizes network bandwidth and prevents timeout errors during large-volume data transmission. In use cases where field operators send proposals with video feeds from factory areas with unstable network environments using smartphones, this system prevents data loss and achieves reliable data collection.

[0097] Step 2: Multi-angle Automatic Evaluation Based on Generative AI. The evaluation index automatic calculation department uses generative AI to analyze the submitted creative descriptions or supplementary materials from multiple perspectives. Generative AI extracts the main points or technical features of the proposal through natural language processing, specifically based on keywords or logical structures such as AI-driven motion optimization, sensor data utilization, and safety enhancement strategies. Combined with image analysis or time series data analysis, it automatically extracts changes in motion patterns, signs of anomalies, the relative positional relationship between operators and robots, bottlenecks within the process, robot operating rate trends, abnormal stop frequency, and interference between operator movement and robot movements from supplementary on-site videos or sensor logs. Furthermore, it compares these findings with internal and external knowledge bases, past improvement cases, industry technology trends, patent information, academic papers, on-site application performance data, and external safety specifications. Each perspective is scored out of 10 based on its innovativeness, feasibility, impact on productivity and quality, and safety, automatically generating the basis for each perspective. As a specific example of a prompt for generative AI to read, one could be: "Analyze the following robot improvement proposal, evaluating it on a scale of 10 out of 10 from four perspectives: innovation, feasibility, impact on productivity and quality, and safety. Calculate the overall evaluation score along with the supporting evidence for each perspective. Proposal description: 'Improve the efficiency and safety of sorting operations by optimizing the robot's movement patterns in collaboration with operators through AI. Attached are sensor logs and on-site video.'" Specifically, in this step, the evaluation index automatic calculation unit initiates an inference pipeline running on a GPU cluster. The system provides input data to the AI ​​model through batch or stream processing. The input data to this AI model includes text tag sequences, image feature vectors, and time-series sensor data. This AI model performs forward propagation calculations on a neural network with hundreds of millions of parameters, extracting the feature representation of the input data. The output data from this AI model includes the scores for each evaluation perspective, along with the supporting text. This output result is used in subsequent processing. The processing unit calculates a weighted average of the scores to calculate the overall evaluation score and updates the evaluation table in the database. To accelerate inference processing, this system utilizes optimized libraries for model quantization and tensor operations. This architecture leverages the technical efficiency of completing the analysis of complex multimodal data within milliseconds. In use cases where hundreds of improvement proposals are submitted daily, and objective scores are assigned instantly without human review, this system maximizes the throughput of the evaluation process.

[0098] Step 3: Integration of Business Performance Data and Calculation of Overall Contribution. The Business Performance Integration Department automatically collects a wide variety of performance data from on-site employees or engineers, including productivity indicators (processing per hour, in-process work efficiency), quality management scores (defect rate, in-process inspection pass rate), safety compliance rate (accident-free days, number of alarm reports), past awards (business improvement awards, quality improvement awards, safety awards), skills enhancement training and learning records (AI technology training, robot maintenance workshops, on-site safety education), on-site leadership experience, participation in on-site improvement activities, on-site communication scores, on-site peer evaluations, inter-site transfer records, and on-site multi-skilling, etc., and integrates this data with AI-based creativity evaluation scores to calculate the overall contribution. The weighted model can be customized according to on-site or organizational policies and is automatically adjusted based on evaluation policies such as innovation-oriented, safety-oriented, quality-oriented, and on-site contribution-oriented. Specifically, in this step, the Business Performance Integration Department executes an extraction, transformation, and loading pipeline to aggregate information from disparate data sources. This system performs outlier removal using statistical methods or imputation of missing values ​​using an average imputation algorithm on the collected raw data. The preprocessed data is then converted into a multi-dimensional numerical matrix and expanded in memory. The processing unit reads a weighted vector based on the set evaluation criteria from the database and performs an inner product operation with the data matrix. Through this matrix operation, the system simultaneously and rapidly calculates the overall contribution score for all employees. This configuration leverages parallel processing by utilizing the CPU's SIMD instructions to improve processing speed, eliminating sequential calculations based on loops. In cases where the system instantly aggregates performance data from thousands of employees and determines evaluation results by the end of the month's evaluation deadline, the batch processing execution time is significantly reduced.

[0099] Step 4: Visualization and On-site Dashboard Display of Evaluation and Contribution. The visualization department displays evaluation and performance data in various formats on the dashboard, providing evaluation trend charts for each proposal, rankings of creative submissions by site and by operator, heatmaps of overall contribution, radar charts by evaluation item, Gantt charts of improvement proposal progress, innovation trend analysis charts for the entire site, safety score trends for each site, visualization of communication networks within the site, creative flow charts between sites, distribution of awards and honors for each site, skill charts for each site, and many other diverse visualization modes. Managers can clearly understand at a glance which proposals are most prevalent in which site, which operator has contributed to innovation, which evaluation item is weaker in the organization, the knowledge flow between sites, and the maturity of the safety culture in each site. Specifically, in this step, the visualization department distributes the calculated overall contribution and evaluation data to the front-end application. When the system detects data updates, it pushes only the differential data to the client terminal via a WebSocket connection. The browser on the client terminal updates the virtual DOM tree based on the received JSON data, executing a differential rendering algorithm that minimizes the application of the actual DOM. In rendering complex graphics, this terminal utilizes WebWorker background processing to avoid blocking the main thread. This configuration effectively prevents browser freezing when rendering graphics containing large amounts of data, providing a smooth user interface. In use cases where the system is constantly displayed on a dashboard on a large on-site monitor, providing real-time and visual feedback on the latest innovations, this powerfully supports on-site motivation enhancement and data-driven decision-making.

[0100] Step 5: Optimal Compensation and Recognition Design Based on Generative AI. The Compensation Design Department uses generative AI to perform multivariate analysis on factors such as overall contribution, past recognition history, employee or engineer motivation, on-site compensation policy, budget status, career aspirations, past incentive responses, on-site leadership experience, on-site improvement activity participation history, on-site talent allocation, and the distribution of compensation history for each site. This automatically designs a variety of compensation and recognition plans corresponding to the site's or individual's aspirations, including special recognitions, bonuses, promotion to on-site improvement project leader, participation in joint development with robot manufacturers, skill certification points, participation in skills enhancement training, introductions in company publicity, invitations to on-site activities, inter-site transfer recommendations, granting new project initiation authority within the site, and on-site mentor role recommendations. Specific examples of prompts for the generative AI to read include: "Considering employee X's overall contribution, past recognition history, motivation, and on-site compensation policy, design the optimal compensation and recognition plan. Explain the proposal content along with the supporting evidence." Specifically, the Compensation Design Department receives employee attribute data and evaluation data as input. The compensation design department transforms the input data into multi-dimensional vectors. These vectors contain, for example, floating-point values ​​representing overall contribution, one-hot encoded arrays representing past awards and recognitions, and embedding vectors of natural language text representing motivational tendencies. The compensation design department maintains a hybrid neural network combining a large-scale language model and a multilayer perceptron. This hybrid neural network employs a Transformer architecture with a self-attention mechanism. The compensation design department inputs the multi-dimensional vectors into this hybrid neural network. This hybrid neural network maps the input vectors to a high-dimensional feature space. Through nonlinear operations within the feature space, this hybrid neural network calculates the probability distribution of the optimal compensation plan. As output data, this hybrid neural network generates an array of scores for multiple compensation candidates. The output data includes, for example, a score of 0.85 representing a recommendation for special recognition, a score of 0.92 representing a recommendation for bonus conferment, and a score of 0.76 representing a recommendation for project leader promotion. The compensation design department applies a Softmax function to the output score array. The compensation design department extracts compensation plans with scores exceeding a threshold as final candidates. The compensation design department sends the extracted compensation plans to the subsequent approval workflow processing module. This system achieves objective and highly accurate reward design independent of human subjectivity. Through parallel processing of multivariate analysis, it significantly reduces computation time. This system results in increased employee motivation and reduced turnover. This is because it accurately captures an individual's potential career aspirations in a high-dimensional space and dynamically allocates optimal incentives. This technology can be applied to use cases such as personnel evaluation systems in large-scale manufacturing industries or performance management systems for software development teams.This system also incorporates reinforcement learning, using past performance changes after being rewarded as reward signals to continuously update the model's weights. The system uses a cross-entropy error function to minimize the learning loss. The system optimizes the network parameters using gradient descent. This system is not merely about improving operational efficiency, but rather about improving the internal data processing structure of the computer and addressing the technical challenge of optimizing dynamic resource allocation.

[0101] Step 6: Generative AI-based Feedback and Improvement Suggestion Generation. Based on the AI's evaluation results, the Feedback Generation Department automatically generates action mode optimization solutions. These solutions include: improving feasibility through more detailed analysis of on-site sensor data; recommending additional risk assessments to enhance safety evaluations; effectively collecting additional learning data to improve the accuracy of anomaly detection algorithms; maximizing the effectiveness of energy-saving operation proposals through coordination with on-site work scheduling; reducing installation costs through collaboration with other departments on-site; and expecting safety improvements through coordination with on-site operator training programs. These specific improvement suggestions or action plans are then distributed to operators or engineers. A specific example of a prompt for the generative AI to read could be: "Based on the following evaluation results, generate specific improvement suggestions and future action plans for employees. Evaluation results: Innovation 9 points, Feasibility 6 points, Productivity and Quality Impact 8 points, Safety 7 points." Specifically, this Feedback Generation Department obtains the evaluation score array output from the evaluation index automatic calculation department and the text data. This Feedback Generation Department then converts the obtained data into a structured prompt tensor. The structured cue tensor contains numerical data such as innovation scores, marker data indicating low likelihood of implementation, and text embedding vectors related to security. This feedback generation unit maintains a large-scale language model specialized for natural language generation. This large-scale language model is a pre-trained Transformer model with billions of parameters. The feedback generation unit inputs the structured cue tensor into this large-scale language model. This large-scale language model processes the input tensor using a decoder network. This large-scale language model successively calculates the marker probability distribution corresponding to the context. As output data, this large-scale language model outputs natural language text representing specific improvement actions and priority scores. The output data includes, for example, text prompting additional analysis of sensor data, itemized text representing implementation steps for risk assessment, and text representing specific methods for learning data collection. The feedback generation unit inputs the output text data into the syntax analysis module. This syntax analysis module extracts specific task items from the text. The feedback generation unit registers the extracted task items in a subsequent task management database. Through this configuration, the system fully automates the feedback generation process for proposers. This system homogenizes the quality of feedback, eliminating human bias. This system achieves the technical effect of repeatedly increasing the likelihood of proposals being implemented. This is because, for projects with low evaluation scores, the system dynamically combines the best improvement strategies extracted from a database of past success stories and provides suggestions. This technology can be applied to use cases such as improvement proposal systems in manufacturing plants or automated code review systems in software development. The system utilizes the proposer's response history to the generated feedback as a feedback loop. The system also includes variations that fine-tune the model to generate text that encourages more behavioral changes by adjusting the reward function of reinforcement learning.This system dynamically changes the parameters of the beam search algorithm during text generation to produce diverse representations. This system does not replace human mental activity, but rather represents an improvement in computer technology by generating novel information structures through probabilistic computations in a high-dimensional language space.

[0102] Step 7: Progress Reporting and Impact Prediction Based on Generative AI. The progress reporting department uses AI to analyze the evaluation status of proposals or the progress and business impact predictions of on-site implementation. For example, the action optimization proposal is currently in the on-site testing phase, with initial evaluations expecting a 15% increase in sorting efficiency and a 20% reduction in safety incident rate. The anomaly detection AI implementation plan requires further discussion due to technical challenges related to residual system integration. The energy-saving operation proposal is expected to reduce monthly energy consumption by 10% and is being implemented simultaneously on multiple production lines on-site. The effects between production lines are automatically summarized. Safety improvement measures linked to on-site operator training programs are underway. Quantitative progress and impact predictions are then distributed as reports to relevant personnel. A specific example of a prompt for the generative AI to read could be, "Based on the progress status and evaluation history of the following ideas, predict future business impact and generate a progress report for relevant personnel." Specifically, this progress reporting department obtains time-series start-up log data and project management data from multiple on-site systems. The department preprocesses the acquired data into time-series tensor data. The time-series tensor data includes, for example, a time-series array of sorting operation times over the past 30 days, a time-series vector of the number of safety incidents, and continuous data on energy consumption. This progress reporting department maintains a prediction model that combines Long Short-Term Memory (LSTM) networks and graph convolutional networks. This progress reporting department inputs the time-series tensor data into this prediction model. This prediction model uses past trends to regress and predict future performance indicators. As output data, this prediction model outputs predicted values ​​and confidence intervals for specific future time points. The output data includes, for example, values ​​representing a 15% increase in operational efficiency after 3 months, a 20% reduction in incidents, and a 10% reduction in energy consumption. This progress reporting department inputs these predicted values ​​into a large-scale language model. This large-scale language model generates natural language report text based on the predicted values. This progress reporting department forwards the generated report text to a subsequent distribution server. This distribution server pushes notification reports to the terminals of relevant personnel. Through this structure, this system achieves real-time, data-driven progress management. This system detects project delay risks early and automates the reallocation of resources. This system delivers a significant improvement in the accuracy of business impact predictions. This is because it doesn't rely on simple linear interpolation, but rather uses graph neural networks to model the nonlinear interactions of multiple related variables. This technology can be applied to use cases such as operational management systems in large-scale factories or monitoring the progress of infrastructure development projects in smart cities. The system also includes modifications such as calculating prediction errors using a mean squared error function and periodically updating the prediction model weights using backpropagation. To ensure robust predictions even with input data containing outliers, the system applies Dropout processing to the input layer.This system is not simply a summary of information, but rather performs a computer-specific processing function: calculating the future state based on dynamic analysis of high-dimensional time series data.

[0103] (specific use case) For example, imagine a scenario where an employee on a factory assembly line submits a proposal from a terminal suggesting using AI to optimize the motion patterns of collaborative robots to improve the efficiency and safety of sorting operations. The idea receiving department receives the proposal titled "Improving Sorting Efficiency Through Collaborative Robot Motion Optimization," with detailed explanations such as "In the current sorting operations, the robot's motion patterns partially overlap with the operator's movement, leading to reduced efficiency or safety risks. By importing an AI-based motion pattern optimization algorithm and analyzing sensor logs and on-site video, the algorithm can automatically optimize the operator's movement and the robot's motion, simultaneously improving efficiency and safety." It also includes relevant documentation such as an on-site layout diagram, operation video, sensor logs, simulation results, submitter information, submission date and time, the relevant project name "Assembly Line DX," the projected effects ("10% reduction in sorting cycle time, 30% reduction in safety incident rate"), and the question of "Can it be integrated with the existing on-site control system?" This information is recorded in the server-side database. The receiving department automatically retrieves similar previously submitted ideas such as "Automating Sorting Operations" or "Improving Robot Motion Safety" and provides them to the employee as reference information. Specifically, this creative receiving unit receives multimodal data packets sent from the terminal. These multimodal data packets are composite data structures containing text data, image data, video data, and time-series numerical data. The creative receiving unit parses the received data packets and performs preprocessing corresponding to each data type. The creative receiving unit converts the text data into word embedding vectors. The creative receiving unit inputs the image data of the site layout diagram into a convolutional neural network to extract spatial feature tensors. The creative receiving unit inputs the operation video data into a 3D convolutional neural network to extract spatiotemporal feature tensors. The creative receiving unit normalizes the sensor log data and converts it into time-series array data. The creative receiving unit combines these feature tensors to generate a single high-dimensional proposal vector. The creative receiving unit stores the generated high-dimensional proposal vector in a vector database. The creative receiving unit executes a similarity retrieval algorithm. The creative receiving unit calculates the cosine similarity between the input high-dimensional proposal vector and existing creative vectors in the database. The creative receiving unit extracts existing creatives with cosine similarity exceeding a specified threshold as similar creatives. The creative receiving unit returns the metadata of the extracted similar creatives to the terminal. This system, through its architecture, processes input data in multiple formats within a unified vector space. It accurately extracts past proposals with potential similarities that cannot be detected through text retrieval. This system achieves technical benefits such as preventing duplicate proposals and promoting the reuse of existing knowledge. This is because, by using multimodal feature representations, the system can compare not only the surface keywords of proposals but also their essential technical structure. This technology can be applied to use cases such as R&D knowledge management systems in manufacturing and patent prior art investigation systems.This system also includes a variation that uses an approximate nearest neighbor search algorithm to complete retrieval within milliseconds, even for databases containing millions of items. This system is not simply for data storage; it performs computer-specific processing, including mapping multimodal data to a high-dimensional space and high-speed geometric operations.

[0104] The evaluation index automatic calculation department uses generative AI to analyze the submitted explanatory documents or supplementary materials. Generative AI extracts keywords such as "AI-based motion optimization," "sensor data utilization," and "safety enhancement" through natural language processing. It automatically extracts overlapping areas of operator and robot movement paths, operational rate trends, and abnormal stop frequencies from on-site videos or sensor logs. Furthermore, by comparing these findings with internal and external knowledge bases, past improvement cases, industry technology trends, patent information, academic papers, on-site application performance data, and external safety standards, it scores each viewpoint out of 10 based on innovation, feasibility, impact on productivity and quality, and safety, automatically generating the basis for each viewpoint. For example, it provides specific analysis results such as "Compared to existing collaborative robot application cases, the motion pattern optimization method is novel, and the safety enhancement measures that utilize on-site sensor data are highly original," "The on-site robot control system has high compatibility with existing AI algorithms, and the implementation cost is within acceptable limits," "Simulation confirmed a 10% reduction in sorting cycle time," and "The risk of interference between operator movement and robot movements has been reduced." As a specific example of a prompt for generative AI to read, one could be: "Analyze the following robot application improvement proposal, evaluating it out of 10 points based on four aspects: innovation, feasibility, impact on productivity and quality, and safety. Calculate the overall evaluation score along with the supporting evidence for each aspect. Proposal description: 'Improve the efficiency and safety of sorting operations by optimizing the collaborative robot's motion patterns using AI. Sensor logs and on-site video are attached.'" Specifically, this evaluation metric automatic calculation unit obtains high-dimensional proposal vectors and raw multimodal data from the idea receiving unit. This evaluation metric automatic calculation unit maintains a visual Transformer model for video analysis. This evaluation metric automatic calculation unit inputs on-site video data into this visual Transformer model. This visual Transformer model calculates the optical flow between video frames, extracting the operator's skeletal coordinate data and the robot's end effector trajectory data. This evaluation metric automatic calculation unit performs spatial cross-determination operations on the extracted coordinate data groups, generating a heatmap tensor for overlapping movement lines. This evaluation metric automatic calculation unit inputs sensor log data into the anomaly detection autoencoder, calculating a time-series vector of abnormal stop frequencies from the reconstruction error. The automatic calculation unit of this evaluation index integrates these analysis results with the proposal text and inputs them into a large-scale language model. This large-scale language model adopts a retrieval-enhanced generative architecture. Based on the input data, this large-scale language model sends queries to external patent databases or academic paper databases to obtain relevant technical document vectors. This large-scale language model compares and contrasts the obtained technical document vectors with the proposal data through a self-attention mechanism. As output data, this large-scale language model outputs score arrays for four evaluation viewpoints and the supporting text.The output data includes numerical arrays such as an innovation score of 8.5, a feasibility score of 7.2, a productivity score of 9.0, and a safety score of 8.8. The automatic calculation unit for these evaluation metrics sends the output score arrays to the subsequent business performance integration unit. Through this structure, the system achieves instantaneous comparison with vast amounts of external knowledge that is impossible for human reviewers. The system automatically extracts quantitative evaluation criteria from unstructured data such as video or sensor data. This system brings about improved objectivity in evaluation and a dramatic acceleration of the review process. This is because the system combines multimodal feature extraction and retrieval enhancement generation, verifying the technical appropriateness of the proposal content based on multi-faceted data sources. This technology can be applied to scenarios such as venture capital investment project evaluation systems or public research grant review systems. The system also includes uncertainty estimation of the evaluation model's output, setting variations of indicators requiring human expert review in cases of low reliability. This system does not mimic human judgment but performs computer-specific data processing based on high-dimensional tensor operations and large-scale vector retrieval.

[0105] The Business Performance Integration Department automatically collects employee productivity metrics (e.g., number of sorts per hour), quality management scores (e.g., defect rate, in-process inspection pass rate), safety compliance rates (e.g., accident-free days, number of near misses reported), past awards and recognitions (e.g., business improvement awards, quality enhancement awards, safety awards), and skills training and workshop attendance records (e.g., AI technology training, robot maintenance workshops, on-site safety training), and integrates these with AI-based creativity evaluation scores to calculate overall contribution. For example, if an employee maintaining high productivity and safety compliance makes an excellent proposal, their overall contribution score increases, and their priority for compensation or recognition rises. Specifically, the Business Performance Integration Department obtains employee performance data from multiple core systems within the company via asynchronous communication. The department normalizes the acquired heterogeneous data to generate a performance feature vector for each employee. This performance feature vector includes, for example, normalized values ​​representing productivity metrics, percentage values ​​representing quality management scores, continuous values ​​representing safety compliance rates, and binary arrays representing award and recognition records. The Business Performance Integration Department receives the creative evaluation score array from the Evaluation Index Automatic Calculation Department. This department combines the performance feature vector with the creative evaluation score array to generate an integrated feature tensor. The department maintains an integrated evaluation model based on a gradient boosting decision tree algorithm. This department inputs the integrated feature tensor into this integrated evaluation model. The integrated evaluation model evaluates the nonlinear interactions of various features through ensemble learning of multiple decision trees. As output data, this integrated evaluation model outputs a comprehensive contribution score ranging from 0 to 100. For example, the output data might be a comprehensive contribution score of 88.5 for employee A and 72.3 for employee B. The Business Performance Integration Department then sends the calculated comprehensive contribution score to the subsequent visualization and compensation design departments. Through this structure, the system integrates scattered personnel and business data into a single evaluation axis. This system eliminates data silos and automates comprehensive talent evaluation. This system provides technical benefits such as ensuring fairness in evaluation and improving computational efficiency. The reason for this is that this system eliminates manual data aggregation and dynamically adjusts the importance of each indicator through weighted optimization based on a machine learning model. This technology can be applied to scenarios such as operator evaluation systems in large-scale call centers or performance management systems for logistics warehouse workers. This system also incorporates a supervised learning transformation in the learning of the integrated evaluation model, using past promotion data or project success rates as target variables. This system has the capability to calculate the importance of features as SHAP values, making the calculation basis of the comprehensive contribution score explainable. This system is not simply a computational application, but rather an improvement in computer technology by providing dynamic feature weighting based on machine learning.

[0106] The Visualization Department displays evaluation and performance data on dashboards, providing features such as evaluation trend charts for "Collaborative Robot Motion Optimization Proposals," rankings of creative submissions by site and by operator, heatmaps of overall contribution, radar charts by evaluation item, and Gantt charts of improvement proposal progress. Managers can easily see which proposals are most prevalent at which sites and which operators have contributed to innovation. Specifically, this Visualization Department acquires various score and forecast data from the Business Performance Integration Department and the Progress Reporting Department. It converts the acquired data into JSON objects for front-end rendering. This Visualization Department has a dynamic rendering engine that runs on a web browser. The Visualization Department inputs the JSON objects into this dynamic rendering engine. This dynamic rendering engine performs coordinate calculations for data points and color space mapping. As output data, this dynamic rendering engine generates interactive graphical user interface elements. Output data includes, for example, SVG data representing line graphs showing changes in evaluation scores relative to a timeline, Canvas data representing two-dimensional heatmaps using color intensity to indicate overall contribution, and vertex coordinate data for radar charts representing multi-dimensional evaluation items using polygons. This visualization unit monitors user input events such as clicks and scrolling. Based on these input events, it performs drill-down processing to dynamically change the granularity of the data to be displayed. Through this configuration, the system transforms complex multidimensional data into intuitive and understandable visual information. The system optimizes the rendering load on the client terminal, rendering large amounts of data points without delay. This system brings technological benefits such as improved decision-making speed for managers and increased efficiency in data exploration. This is because the system separates server-side data aggregation from asynchronous client-side rendering, providing an interactive data exploration environment that responds instantly to user actions. This technology can be applied to scenarios such as business dashboard systems or factory equipment operation monitoring systems. The system also includes a personalized variant that learns user eye movement data or operation history, automatically placing the most frequently referenced charts at the top of the dashboard. This system is not simply a list of information, but an improvement on the computer-specific user interface technology of visual mapping and dynamic reconstruction of multidimensional data.

[0107] The Compensation Design Department uses generative AI to perform multivariate analysis on factors such as overall contribution, past recognition history, employee motivation, on-site compensation policies, budget status, career aspirations, and past incentive responses. This automatically designs optimal compensation and recognition plans that align with the on-site or individual aspirations, such as "special recognition + bonus," "promotion to on-site improvement project leader," "participation in joint development with robot manufacturers," and "skill recognition points." A specific example of a prompt for the generative AI could be, "Considering employee X's overall contribution, past recognition history, motivation, and on-site compensation policies, design the optimal compensation and recognition plan. Explain the proposal content along with the supporting evidence." Specifically, the Compensation Design Department obtains overall contribution scores from the Business Performance Integration Department and employee personal data from the HR database. The department combines this data to generate employee state vectors. These vectors include, for example, normalized values ​​representing overall contribution, remaining balance values ​​representing budget status, and encoded arrays of categorical variables representing past incentive responses. The Compensation Design Department maintains a compensation optimization agent based on deep reinforcement learning algorithms. This compensation design department inputs employee state vectors into a compensation optimization agent. Based on a Markov decision process model, this agent calculates the optimal action value using the current policy network. As output data, the agent outputs an array of action values ​​(Q-values) for multiple compensation plans. Examples of output data include Q-values ​​such as 0.88 for a combination of special recognition and bonuses, 0.95 for a project leader promotion, and 0.72 for collaborative development participation rights. The compensation design department selects the compensation plan with the highest Q-value. The identifier of the selected compensation plan is input into a large language model. This model generates natural language text explaining the legitimacy of the selected compensation plan. The compensation design department then sends the generated text and the compensation plan identifier to the subsequent approval system. Through this structure, the system solves the resource allocation problem of maximizing organizational motivation within a limited budget. This system automates the compensation design process, reducing the workload of managers. This system delivers the technical effect of optimizing the return on investment of incentives. The reason for this is that this system uses past rewards assigned to resumes and subsequent performance changes as reward signals to continuously update the reinforcement learning model, dynamically exploring the rewards most suitable for individual characteristics. This technology can be applied to scenarios such as incentive calculation systems in sales departments or dynamic task reward setting systems in freelance platforms. This system also includes a variation that uses an ε-greedy algorithm to suggest random reward plans with a certain probability to adjust the trade-off between exploration and exploitation, thereby improving the model's generalization performance. This system replaces business operations that rely on human intuition with the computer technology of solving optimization problems in a high-dimensional state space.

[0108] Based on AI-based evaluation results, the Feedback Generation Department automatically generates specific improvement suggestions or action plans, such as "The possibility of implementing the action mode optimization solution is increased by analyzing the sensor data on site in more detail," "To strengthen the safety evaluation, it is recommended to implement an additional risk assessment," "Additional learning data collection is effective in improving the accuracy of the anomaly detection algorithm," and "Energy-saving operation proposals maximize their effectiveness by linking with the on-site operation schedule," and distributes them to operators or engineers. As a specific example of a prompt text that the generative AI can read, it could be: "Based on the following evaluation results, specific improvement suggestions and future action plans are generated for employees. Evaluation results: Innovation 8 points, Feasibility 7 points, Productivity and Quality Impact 9 points, Safety 8 points." Specifically, this Feedback Generation Department obtains the scores for each viewpoint and the extracted topic text from the evaluation indicator automatic calculation department. This Feedback Generation Department formats the obtained data into input prompt strings. The input prompt strings contain values ​​such as an innovation score of 8 points, a feasibility score of 7 points, and a topic marker indicating insufficient sensor data analysis. This Feedback Generation Department maintains a finely tuned generative Transformer model. This feedback generation unit inputs the input prompt string into this generative Transformer model. This generative Transformer model uses a self-attention mechanism to analyze the dependencies between input tokens. As output data, this generative Transformer model autoregressively generates a sequence of text suggestions for improvement. The output data includes, for example, text representing detailed analysis steps of sensor data, text representing checklists for additional risk assessments, and text representing templates for automated scripts for learning data collection. This feedback generation unit inputs the generated text sequence into a natural language processing filter to verify whether it contains inappropriate expressions or content that violates company regulations. This feedback generation unit sends the verified text data to a messaging server. This messaging server asynchronously distributes the message to the terminals of the target operator or engineer. Through this configuration, the system provides immediate constructive feedback on the evaluation results. This system accelerates the proposal refinement cycle. This system brings about the technical effect of continuously improving the quality of ideas. This is because this system does not rely solely on simple score notifications, but uses a model that has learned how similar past proposals were improved and adopted to generate specific next steps. This technology can be applied to scenarios such as automated report grading systems for students or automated code correction suggestion systems for programmers. The system also includes variants with parameter control functionality that dynamically adjusts the length of the generated feedback text based on the recipient's job title or technical skill level. This system does not replace human guidance, but rather leverages the reasoning capabilities of large language models to improve the efficiency and accuracy of information delivery—an improvement in computer technology.

[0109] The progress reporting department uses AI to analyze the evaluation status of proposals, the progress of on-site implementation, and the prediction of business impact. It distributes quantitative progress and effect predictions, such as "The action optimization proposal is currently in the on-site testing phase, and the initial evaluation expects a 10% increase in sorting efficiency and a 30% reduction in the accident rate," and "Due to the issue of residual interaction with the existing on-site control system, further discussion is needed," as reports to relevant personnel. As a specific example of a prompt for generative AI to read, it could be something like, "Based on the progress status and evaluation history of the following ideas, predict future business impact and generate a progress report for relevant personnel." Specifically, this progress reporting department obtains task completion status from the project management database and real-time operational metrics from the on-site IoT gateway. This department integrates the acquired data to construct a progress state tensor. The progress state tensor includes, for example, classification values ​​representing the on-site testing phase, floating-point numbers representing the current value of sorting efficiency, and time-series arrays representing the shift in the accident rate. This progress reporting department maintains a multi-task learning model that integrates recurrent neural networks for time-series forecasting and language models for text generation. This progress reporting department inputs the progress status tensor into this multi-task learning model. This multi-task learning model simultaneously performs future indicator predictions and report text generation while sharing the hidden layer representation. As output data, this multi-task learning model outputs an array of predicted values ​​and natural language report text. The output data includes predicted values ​​such as a 10% increase in work efficiency, a 30% reduction in accidents, and text indicating the existence of system-related issues. This progress reporting department converts the output report text into a PDF or HTML document. The generated document is automatically sent to the email addresses of relevant personnel. Through this structure, the system automatically documents the project's progress status and future predictions. This system eliminates the manual time required for report preparation. This system brings about technical effects such as increased transparency in project management and early detection of risks. This is because the system comprehensively analyzes real-time sensor data and task progress data, and the neural network detects and reports subtle delays or signs of performance degradation that are difficult for human intuition to perceive. This technology can be applied to agile development management systems for software development, or engineering management systems for construction sites, among other application scenarios. This system also includes a variation that sets alarm flags and sends emergency notifications to administrators' terminals when the predicted business impact is lower than the pre-set target value. This system is not simply a data format conversion, but rather a high-level computer processing system that performs integrated analysis of multimodal data and predictive calculations of future states.

[0110] In this way, by integrating insights from on-site employees or engineers with on-site data, the system achieves end-to-end automated evaluation, integration, reward design, feedback, and progress management based on generative AI from multiple perspectives. This results in a system capable of efficiently and fairly promoting on-site-led innovation and continuous improvement in productivity, quality, and safety in the field of factory robotics. Specifically, this system employs a distributed processing architecture combining multiple independent neural network modules via an asynchronous message queue protocol. The system seamlessly transfers high-dimensional tensor data and metadata between modules. From the vectorization of multimodal data in the input layer to the generation of optimized reward plans or feedback text in the output layer, the system fully automates a series of data pipelines. The system persists intermediate data generated in each processing step to a graph database. This graph database is utilized as a knowledge graph to continuously improve the accuracy of subsequent inference tasks. To minimize overall system latency, a portion of the inference processing is offloaded to edge computing devices. Through this architecture, the system integrates vast amounts of on-site data and tacit human knowledge in the digital space. The system algorithmizes the innovation process itself, executing it in a scalable manner. This system delivers technological benefits such as accelerating the knowledge creation cycle across the entire organization and establishing a sustainable competitive advantage. This is because the system not only automates individual tasks but also constructs a closed-loop machine learning system encompassing evaluation, integration, reward design, and feedback, possessing a mechanism for self-organized evolution. This technology is not limited to manufacturing but can be applied to business improvement systems in healthcare institutions, store operation optimization systems in retail, and new service development platforms in financial institutions, across all industry sectors. The system also incorporates federated learning to enable distributed learning of inference models across modules, sharing model variations across multiple locations while protecting privacy. This system transcends being merely a tool supporting human intellectual activity; it functions as a data-driven, autonomous knowledge processing engine, providing a fundamental improvement to computer technology.

[0111] 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.

[0112] Data generation model 58 is what is known as generative AI (Artificial Intelligence).

[0113] One example of data generation model 58 includes ChatGPT (registered trademark) (Internet search).<URL: https: / / openai.com / blog / chatgpt> Generative AI, etc.

[0114] 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.

[0115] 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.

[0116] 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.

[0117] [Second Implementation] Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0118] 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.

[0119] 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).

[0120] 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.

[0121] 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.

[0122] 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).

[0123] 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.

[0124] Figure 4 An example of the main functions of the data processing device 12 and the smart glasses 214 is shown. Figure 4 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0125] 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.

[0126] 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).

[0127] 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.

[0128] 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".

[0129] (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.

[0130] (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.

[0131] 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.

[0132] 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.

[0133] 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.

[0134] [Third Implementation] Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0135] 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.

[0136] 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).

[0137] 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.

[0138] 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.

[0139] 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).

[0140] 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.

[0141] Figure 6 An example of the main functions of the data processing device 12 and the head-mounted terminal 314 is shown. Figure 6 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0142] 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.

[0143] 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.

[0144] In the head-mounted terminal 314, the processor 46 performs the receiving and output processing.

[0145] The memory 50 stores a 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 the control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0146] 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 head-mounted terminal 314. In the following description, the data processing device 12 will be referred to as the "server" and the head-mounted terminal 314 will be referred to as the "terminal".

[0147] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0148] (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.

[0149] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the head-mounted terminal 314. In the head-mounted terminal 314, the control unit 46A causes the speaker 240 and the display 343 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.

[0150] 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 head-mounted terminal 314, 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 head-mounted terminal 314. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the head-mounted terminal 314 or external devices, and the head-mounted terminal 314 acquires or collects information required for processing from the data processing device 12 or external devices.

[0151] 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 head-mounted terminal 314.

[0152] [Fourth Implementation] Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0153] like Figure 7 As shown, the data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[0154] 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).

[0155] Robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control object 443. 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, control object 443, and communication I / F 44 are also connected to the bus 52.

[0156] 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.

[0157] 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).

[0158] 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.

[0159] The controlled object 443 includes a display device, LEDs for the eyes, and motors for driving the arms, hands, and feet. The posture and movements of the robot 414 are controlled by controlling the motors for the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. In addition, facial expressions of the robot 414 can also be expressed by controlling the illumination state of the LEDs for the robot 414's eyes.

[0160] Figure 8 An example of the main functions of the data processing device 12 and the robot 414 is shown. For example... Figure 8 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0161] 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.

[0162] 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.

[0163] In robot 414, 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 RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on RAM 48.

[0164] 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 robot 414. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 will be referred to as the "terminal".

[0165] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0166] (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.

[0167] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the controlled object 443 to output the result of the specific processing. The microphone 238 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 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0168] Furthermore, the processing performed by the 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 robot 414, 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 robot 414. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the robot 414 or external devices, and the robot 414 acquires or collects information required for processing from the data processing device 12 or external devices.

[0169] 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 robot 414.

[0170] Furthermore, the emotion-specific model 59, serving as an emotion engine, can determine the user's emotion based on a specific mapping. Specifically, the emotion-specific model 59 can determine the user's emotion based on an emotion graph that serves as a specific mapping (see...). Figure 9 The robot's emotions can be determined by the emotion-specific model 59. In addition, the emotion-specific model 59 can also determine the robot's emotions in the same way, and the specific processing unit 290 can also perform specific processing using the robot's emotions.

[0171] Figure 9This is a diagram representing an emotion map 400 that maps various emotions. In the emotion map 400, emotions are arranged radially from the center in concentric circles. The closer to the center of the concentric circles, the more primitive the emotion is. Further out on the concentric circles, emotions are arranged representing states or actions arising from mood. Emotion is a concept that includes both feelings and mental states. To the left of the concentric circles, emotions generated by reactions occurring in the brain are arranged roughly. To the right of the concentric circles, emotions guided by situational judgments are arranged roughly. Above and below the concentric circles, emotions generated by reactions occurring in the brain and guided by situational judgments are arranged roughly. Furthermore, the emotion of "pleasure" is arranged above the concentric circles, and the emotion of "unpleasantness" is arranged below. Thus, in the emotion map 400, various emotions are mapped according to the structure of emotion generation, while easily generated emotions are mapped nearby.

[0172] These emotions are distributed at the 3 o'clock position on the Emotion Chart 400, and usually fluctuate between peace and unease. In the right half of the Emotion Chart 400, because situational awareness is more dominant than internal feelings, it gives a sense of calm.

[0173] The inner side of the emotion diagram 400 represents the mind, and the outer side of the emotion diagram 400 represents actions. Therefore, the further you go to the outer side of the emotion diagram 400, the more the emotion can be seen (manifested in actions).

[0174] Here, human emotions are based on a balance of various factors such as posture and blood sugar levels. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. In robots, cars, and motorcycles, emotions can also be created based on a balance of factors such as posture and remaining battery power. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. Emotion maps can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Speech Emotion Recognition and Brain Physiological Signal Analysis Systems for Emotion, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). In the left half of the emotion map, emotions belonging to the "Reaction" domain, where sensation is dominant, are arranged. Furthermore, in the right half of the emotion map, emotions belonging to the "Situation" domain, where situational cognition is dominant, are arranged.

[0175] The emotion map defines two types of emotions that promote learning. One is a negative emotion located near the middle of "repentance" or "reflection" on the situation side. That is, when the robot experiences negative emotions such as "I never want to feel this way again" or "I never want to be scolded again." The other is a positive emotion located near "desire" on the response side. That is, when the robot experiences positive feelings such as "wanting more" or "wanting to know more."

[0176] The emotion-specific model 59 feeds user input into a pre-learned neural network to obtain emotion values ​​representing each emotion shown in the emotion graph 400, and determines the user's emotion. This neural network is pre-learned based on multiple learning data sets that combine user input with emotion values ​​representing each emotion shown in the emotion graph 400. Furthermore, this neural network is learned to... Figure 10 As shown in sentiment graph 900, nearby sentiment values ​​are similar to each other. Figure 10 Examples show how emotions such as "peace of mind," "stability," and "reassurance" can result in similar emotional values.

[0177] The above description primarily focuses on the functions of the data processing device 12 to illustrate the system disclosed herein. However, the system disclosed herein may not necessarily be implemented on a server. The system disclosed herein can be implemented as a general information processing system. For example, it can be implemented as a software program running on a personal computer, an application running on a smartphone, etc. The methods disclosed herein can be provided to users in the form of SaaS (Software as a Service).

[0178] In the above embodiments, an example of specific processing performed by a single computer 22 is given. However, the technology disclosed herein is not limited to this, and distributed processing of specific processes by multiple computers, including computer 22, is also possible. For example, the data generation model 58 may also be provided in an external device of the data processing apparatus 12, where data is generated based on input data.

[0179] In the above embodiments, an example of storing a specific processing program 56 in memory 32 is illustrated, but the technology disclosed herein is not limited thereto. For example, the specific processing program 56 may also be stored in a portable computer-readable non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 performs specific processing according to the specific processing program 56.

[0180] Alternatively, the specific processing program 56 can be stored in a storage device such as a server connected to the data processing device 12 via a network 54, and the specific processing program 56 can be downloaded and installed into the computer 22 upon request from the data processing device 12.

[0181] Furthermore, it is not necessary to store the entire specific process 56 in a storage device such as a server connected to the data processing device 12 via the network 54, nor is it necessary to store the entire specific process 56 in the memory 32; a portion of the specific process 56 may also be stored.

[0182] As a hardware resource for performing specific processing, various processors can be used. For example, a CPU can be listed as a processor; it is a general-purpose processor that functions as a hardware resource for performing specific processing by executing software, i.e., a program. Furthermore, as a processor, a dedicated circuit can be listed as a processor; it is a processor with a circuit structure specifically designed for performing specific processing, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit).

[0183] Every processor has built-in or connected memory, and every processor performs specific processes by using memory.

[0184] The hardware resources for performing a specific process can consist of one of these various processors, or a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Furthermore, the hardware resources for performing a specific process can also be a single processor.

[0185] As an example of a single processor, the first type consists of a combination of one or more CPUs and software, which functions as a hardware resource to perform specific processing. The second type uses a processor, such as a System-on-a-chip (SoC), which implements the entire system functionality, including multiple hardware resources performing specific processing, using a single IC chip. In this case, the specific processing is implemented using one or more of the aforementioned processors that serve as hardware resources.

[0186] Furthermore, as the hardware architecture of these various processors, more specifically, circuits combining semiconductor elements and other circuit components can be used. Moreover, the specific process described above is merely an example. Therefore, it goes without saying that, without departing from the main point, unnecessary steps can be removed, new steps can be added, or the processing order can be changed.

[0187] The foregoing descriptions and illustrations are detailed explanations of the parts covered by this disclosure and are merely one example of this disclosure. For instance, the descriptions of the above-described structure, function, role, and effect are just one example of the structure, function, role, and effect of the parts covered by this disclosure. Therefore, it goes without saying that, without departing from the spirit of this disclosure, unnecessary parts can be deleted, new elements can be added, or replacements can be made to the foregoing descriptions and illustrations. Furthermore, to avoid confusion and facilitate understanding of the parts covered by this disclosure, explanations of technical common sense that does not require special explanation for implementing this disclosure have been omitted from the foregoing descriptions and illustrations.

[0188] All documents, patent applications and technical standards described in this specification are incorporated herein by reference as if they were specifically and individually described as incorporated by reference.

[0189] In addition, the following notes further explain the above explanation.

[0190] (Note 1) A system for evaluating and managing new proposals or improvement ideas submitted by field employees or engineers in the field of factory robotics includes an idea receiving department, an automatic evaluation index calculation department, a performance integration department, a visualization display department, a compensation design department, a feedback generation department, and a progress reporting department. The system is characterized by receiving ideas from field employees or engineers via terminals, including titles, detailed descriptions, related materials, sensor logs, operation videos, field drawings, submitter information, submission date and time, related project names, and envisioned effects, and recording these in a server-side database. The automatic evaluation index calculation department uses generative AI, through natural language processing or knowledge base comparison, to compare the content or additional data of the received ideas with past similar cases, technological trends, patent information, academic papers, and field data, automatically calculating various evaluation indicators such as innovativeness, feasibility, impact on productivity and quality, and safety. The performance integration department... The Performance Integration Department automatically collects performance data such as productivity indicators, quality management scores, safety compliance rates, commendation records, and skills enhancement training records of on-site employees or engineers, and integrates them with the evaluation indicators to calculate the overall contribution. The Visualization Display Department displays the evaluation indicators and overall contribution in various forms such as charts, heat maps, radar charts, and progress Gantt charts on the dashboard. The Compensation Design Department uses generative AI to analyze the overall contribution, past commendation records, motivational tendencies, on-site compensation policies, budget status, and career aspirations, and automatically designs the optimal compensation and commendation plan corresponding to the on-site or individual aspirations. The Feedback Generation Department uses generative AI to automatically generate evaluation results, improvement suggestions, and specific action plans, and distributes them to employees or engineers. The Progress Reporting Department uses generative AI to analyze the evaluation status of proposals, the progress of on-site implementation, and business impact predictions, and distributes quantitative progress and effect predictions as reports to relevant personnel.

[0191] (Note 2) As described in Appendix 1, the system is characterized in that the automatic evaluation index calculation unit uses generative AI to analyze the submitted robot application and improvement proposal descriptions or additional sensor logs, operation videos, on-site drawings, etc., and combines natural language processing, image analysis, and time series data analysis to score each evaluation index of innovation, feasibility, productivity and quality impact, and safety out of 10 points, automatically generating the basis for each viewpoint, and calculating the comprehensive evaluation score by comparing it with similar past proposals, industry technology trends, patent information, and on-site application performance data, and visually displaying it on the dashboard along with the evaluation basis.

[0192] (Note 3) As described in Appendix 1, the system is characterized in that the compensation design department uses generative AI to perform multivariate analysis on comprehensive contribution, past commendation history, employee or engineer motivation, on-site compensation policy, budget status, career aspirations, and past incentive responses, optimizing and automatically designing various compensation and commendation plans for each employee or engineer, such as special commendations, bonuses, promotion of on-site improvement project leaders, participation in joint development with robot manufacturers, and skill certification points; furthermore, the feedback generation department uses generative AI to automatically generate evaluation results, improvement suggestions, and specific action plans for on-site implementation; and the progress reporting department uses generative AI to analyze the evaluation status of proposals, the progress of on-site implementation, and business impact predictions, automatically distributing reports containing quantitative progress and effect predictions to relevant personnel.

Claims

1. A creative evaluation system using generative AI, comprising a creative receiving department, an automatic evaluation index calculation department, a business performance integration department, a visualization display department, a compensation design department, a feedback generation department, and a progress report department, characterized in that, The Creative Receiving Department receives new creative submissions from employees, including content, submitter information, submission date and time, and records them in the database. The Evaluation Indicator Automatic Calculation Department uses generative AI to automatically calculate multiple evaluation indicators such as the innovativeness, feasibility, and economic impact of the received creatives through natural language processing or knowledge base analysis. The Business Performance Integration Department integrates the employees' typical business performance data with the evaluation indicators to calculate the overall contribution. The Visualization Display Department displays the evaluation indicators and overall contribution in visual form such as charts or tables on a dashboard. The Compensation Design Department uses generative AI to automatically design the optimal compensation or recognition content for each employee based on the overall contribution. The Feedback Generation Department uses generative AI to automatically generate evaluation results or improvement suggestions and provide them to employees or supervisors. The Progress Reporting Department uses generative AI to automatically report the evaluation status or progress of creatives, as well as predicted impacts on business, to relevant personnel.

2. The system as described in claim 1, characterized in that, The automatic evaluation index calculation unit uses generative AI to analyze the content of submitted ideas through natural language processing. By comparing it with knowledge bases inside and outside the company or with similar past case data, it automatically calculates standardized scores for each evaluation index, including innovation, feasibility, and economic impact. The scores are then weighted to calculate a comprehensive evaluation score. The visualization display unit displays various evaluation data, including the comprehensive evaluation score, as charts or tables on the dashboard in real time.

3. The system as described in claim 1, characterized in that, The compensation design department uses generative AI to analyze the overall contribution and the employee's past recognition history or motivation data to automatically determine the optimized personalized compensation or recognition content for each employee. Furthermore, the feedback generation department uses generative AI to automatically generate specific improvement suggestions or future action plans corresponding to the evaluation results and provide them to employees and their supervisors. The progress reporting department uses generative AI to automatically analyze the progress status of submitted ideas or predict their business impact, and automatically distributes these reports as real-time reports that all relevant personnel can view.

Images