Learning management system for online students

Abstract

By prioritizing and integrating contexts across three levels—organization, location, and individual—to accurately identify user intent, and by physically restricting AI-driven data access, we provide a highly accurate and secure business support system. [Solution] In an online student learning management system 1 comprising a user terminal 101 used by online students connected via a network and an information processing server 103, the information processing server 103 is characterized by having a ToDo optimization management unit that optimizes and manages ToDos on a daily basis, an automatic curriculum generation unit that automatically generates a curriculum until goal achievement, and a workflow integrated AI generation unit that uses generation AI to integrate a workflow optimized for generating learning deliverables.

Description

[[Technical Field]]

[0001] The present invention relates to a learning management system for online students, and to an information processing system that provides business support using artificial intelligence (AI) in environments with hierarchical organizational structures such as educational institutions, companies, and research institutions. In particular, the present invention relates to a system architecture that achieves both context understanding for each organizational hierarchy and strict data access control. [[Background Art]]

[0002] Conventionally, in organizations such as cram schools and companies, grade management, customer management, daily business reports and the like have been databased. In recent years, attempts have been made to utilize such data using generative AI (LLM), but general AI models cannot understand "organization-specific technical terms" and "local rules for each base". For example, whether an instruction "keep an eye on the situation" refers to grades (quantitative data) or mental state (qualitative data) varies depending on the attributes of the speaker and the customs of the base.

[0003] Furthermore, in a multi-tenant environment (a system where multiple organizations and bases coexist), granting AI agents extensive data access rights for answer generation entails the security risk of information leakage due to hallucination, where the agent references data or confidential information from other bases that it originally has no access rights to.

[0004] Patent Document 1 discloses a system for supporting reskilling for middle-to-senior age groups utilizing generative AI, the system comprising digital skill training means, curriculum creation means tailored to user needs, and learning progress management means.

[0005] Patent Document 2 discloses an online web learning system utilizing generative AI, the system comprising means for creating a curriculum based on a user's questions and answers and generating questions according to the situation, and a web service accessible 24 hours a day, 365 days a year.

[0006] Patent Document 3 discloses "a system that includes means for formulating an educational curriculum that takes into account learning needs, goals, behavioral characteristics and aptitudes, and diverse backgrounds (external environments such as home environment), means for visualizing what needs to be done to achieve the goals by clarifying evaluation methods and progress management, and means for sharing with teachers, allowing teachers to check the progress of the curriculum formulated by AI and provide appropriate support such as motivation." [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2025-058334 [Patent Document 2] Japanese Patent Publication No. 2025-055329 [Patent Document 3] Japanese Patent Publication No. 2025-049989 [Overview of the Initiative] [Problems that the invention aims to solve]

[0008] Patent Document 1 discloses a system for providing reskilling support for middle-aged and senior individuals using generative AI, which includes a digital skills training method, a curriculum creation method tailored to user needs, and a learning progress management method. However, it did not consider integrating contexts across three levels—organization, location, and individual—with priority given to accurately identifying the user's intent.

[0009] Patent Document 2 discloses an online web learning system that utilizes generative AI to create a curriculum based on user questions and answers, generates problems and other elements according to the situation, and includes a web service that can be accessed 24 hours a day, 365 days a year. However, it did not consider integrating the context across three levels—organization, location, and individual—with priority given to accurately identify the user's intent.

[0010] Patent Document 3 discloses a system that includes means for formulating educational curricula that take into account learning needs, goals, behavioral characteristics and aptitudes, and diverse backgrounds (external environments such as home environment), means for visualizing what needs to be done to achieve goals by clarifying evaluation methods and progress management, and means for sharing with teachers, allowing teachers to check the progress of the curriculum formulated by AI and provide appropriate motivation and other support. However, it did not take into consideration the integration of contexts across three levels—organization, location, and individual—with priority given to accurately identifying the user's intent.

[0011] Furthermore, while Patent Documents 1-3 share the common feature of utilizing generative AI in online web-based learning, they did not consider integrating contexts across three levels—organization, location, and individual—with prioritization, nor did they consider accurately identifying the user's intent. As a result, learners could proceed with their learning by referring to the high-quality generated deliverables as a "standard for correct answers," but the system did not enable simultaneous connection between theory and practice, thus not allowing for the fastest skill acquisition. The system has not yet reached a level where the intended learning effect can be proven with evidence.

[0012] This invention has been made in view of the above problems, and aims to provide a highly accurate and secure online learning management system for students by prioritizing and integrating contexts across three levels—organization, location, and individual—to accurately identify user intent, and by confining AI-driven data access to a physically restricted area.

[0013] Another objective of the present invention is to provide a highly accurate and secure business support system by prioritizing and integrating contexts across three levels—organization, location, and individual—to accurately identify user intent, while also confining AI-driven data access to a physically restricted area. [Means for solving the problem]

[0014] The learning management system for online students according to the first invention of the present invention is a learning management system for online students comprising a user terminal used by online students connected via a network and an information processing server, wherein the information processing server comprises a ToDo optimization management unit that optimizes and manages ToDos on a daily basis, an automatic curriculum generation unit that automatically generates a curriculum until goal achievement, and a workflow integrated AI generation unit that uses generation AI to integrate a workflow optimized for generating learning deliverables. The system includes a context generation unit that generates user-specific contextual information by integrating three levels of profiles—organization, location, and individual user—in a predetermined order of priority, and an evidence pack generation unit that physically pre-filters search target data using metadata associated with the organization and location, and generates an evidence pack containing only the confirmed facts extracted by the pre-filtering. The generating AI generates natural language responses based only on the confirmed facts stored in the evidence pack. It is characterized by the following:

[0015] The learning management system for online students according to the second invention of the present invention is characterized in that, in the first invention, the ToDo optimization management unit includes a notation step of breaking down the individual curriculum generated by the automatic curriculum generation unit into daily tasks and displaying "What to do today" on the individual top page of the online student; a transmission step of sending a reminder message to the online student's user terminal via the official SNS if the daily ToDo tasks displayed in the notation step are not completed by an arbitrarily determined time; and a reflection step of defining the progress status at 23:59 each day as "current progress," automatically recalculating the goal achievement curriculum for the following day and beyond, and reflecting this in the display of the latest version of the daily ToDo using the same method as the notation step.

[0016] The learning management system for online students according to the third invention of the present invention is characterized in that, in the second invention, the automatic curriculum generation unit comprises: an interview step in which the automatic curriculum generation unit conducts an interview with the online student in a dialogue format, focusing on items such as available time (weekdays and holidays), target results, current skill level, and skills and work content that the student has worked on and / or learned in the past, to understand the student's current situation and goals; and a curriculum generation step in which the curriculum for the student is generated from the learning progress and achievement results database of online students collected by the ToDo optimization management unit, based on the degree of match with past online student cases.

[0017] The learning management system for online students according to the fourth invention of the present invention, in the third invention, the workflow-integrated AI generation unit is Based on the selection input of a specific business domain from the user terminal, a workflow template that defines the process for generating deliverables, which is stored in the storage of the information processing server in advance and associated with that specific business domain, is generated. A selection step to select, and based on the workflow selected in the selection step, The generation The system is characterized by comprising: an extraction step in which the AI ​​conducts dynamic interviews with the online students and extracts elements necessary for generating deliverables; and an AI generation step in which the AI ​​generates everything from strategic design to specific deliverables based on the elements extracted in the extraction step.

[0018] The learning management system for online students according to the fifth invention of the present invention is characterized in that, in the extraction step of the fourth invention, a proprietary inference model that integrates the know-how of the instructor and expert team or a highly tuned set of prompts is executed in the backend.

[0019] The learning management system for online students according to the sixth invention of the present invention further comprises, in the first invention, an information processing server, a receiving unit that receives queries input from the user terminal, an intent determination unit that interprets the intent of the query and determines whether the query is an inquiry involving statistical processing or calculation of specific data, a tool execution unit that, if the intent determination unit determines that it is an inquiry involving calculation, bypasses the document text reading and semantic interpretation process by the generating AI and performs a direct search process or calculation process on an external database, and means for storing definitive numerical values ​​or extracted results obtained from processing by the external database as a set of premise facts (evidence pack) for answer generation, wherein the generating AI generates an answer in natural language based only on the definitive facts stored in the evidence pack.

[0020] A learning management system for online students according to a seventh aspect of the present invention is characterized in that, in the sixth aspect, the information processing server further comprises a tag management unit that assigns metadata (tags) linked to an organization identifier and a base identifier to student, instructor, teaching material, and comment data, and when generating the evidence pack, prior to performing similarity search (vector search) by the generative AI, the data to be searched is physically pre-filtered using the metadata, thereby eliminating mixing of irrelevant data. Effects of the Invention

[0021] According to the present invention having the above configuration, contexts across three hierarchies of organization, base, and individual are integrated with prioritization to accurately identify a user's intention, and AI-based data access is confined within a physically restricted range, whereby a high-accuracy and high-security learning management system for online students can be realized.

[0022] Further, according to the present invention, a student proceeds with learning while referring to the generated high-quality deliverable as a "correct answer standard", thereby connecting theory and practice at the same time, and a learning management system for online students that enables skill acquisition in the shortest time can be realized. Brief Description of the Drawings

[0023] The drawings show specific embodiments of the present invention according to the present disclosure, and include not only the essential configurations of the invention, but also selective and preferred embodiments. [Figure 1] Fig. 1 is an overall configuration diagram of a business support system according to an embodiment of the present invention. [Figure 2] Fig. 2 is a functional block diagram of an information processing server 103. [Figure 3] Fig. 3 is a conceptual diagram of a data structure and a priority table of a hierarchical profile (UIP / CIP / OIP). [Figure 4] Fig. 4 is a processing flowchart from query reception to evidence pack generation and answer output. [Figure 5] Figure 5 is a sequence diagram of the account integration process using shadow profiles. [Figure 6A] Figure 6A shows an example of the initial screen of a web-based browser on user terminal 101 in the learning management system of this embodiment. [Figure 6B] Figure 6B shows an example of a summary comment for each icon in the right corner of the initial browser screen in Figure 6A, along with an overview of the function of each icon. [Figure 7] Figure 7 shows an example of the student context screen that appears when the student information icon is clicked. [Figure 8] Figure 8 shows an example of the schedule management screen that appears when you click the calendar icon. [Figure 9] Figure 9 shows an example of the student wellness dashboard screen that appears when the pulse icon is clicked. [Figure 10] Figure 10 shows an example of the audio scriber screen that appears when the transcription icon is clicked. [Figure 11] Figure 11 shows an example of the student comment management screen that appears when the comment icon is clicked. [Figure 12] Figure 12 shows an example of the grade management screen that appears when the grade icon is clicked. [Figure 13A] Figure 13A shows an example of the route tracker screen (part 1) that appears when the route icon is clicked. [Figure 13B] Figure 13B shows an example of the route tracker screen (part 2) that appears when the route icon is clicked. [Figure 14] Figure 14 shows an example of the resource management screen that appears when you click on a resource icon. [Figure 15] Figure 15 shows an example of the resource import screen that appears when you click the "Upload Invoice / Inventory" icon in Figure 14. [Figure 16] Figure 16 shows an example of the resource allocation screen that appears when you click the "Allocate Resources" icon in Figure 14. [Figure 17A] Figure 17A shows an example of the management dashboard screen (part 1) that appears when the Value icon is clicked. [Figure 17B] Figure 17B shows an example of the management dashboard screen (part 2) that appears when the Value icon is clicked. [Figure 18] Figure 18 shows an example of the smart transaction registration screen that appears when you click the "Register Transaction" icon in Figure 17A. [Figure 19A] Figure 19A shows an example of the suggestion generator screen that appears when the suggestion icon is clicked. [Figure 19B] Figure 19B shows an example of the AI-generated report display screen that appears when you click the "Start In-Depth Research" button in Figure 19A. [Figure 20] Figure 20 shows an example of the display screen for the proposal slides for parents that appear when the "Convert to Slides" button in Figure 19B is clicked. [Figure 21] Figure 21 shows an example of the display screen for internal materials (for department heads and instructors) that appear when the "Generate internal script" button is clicked. [Figure 22] Figure 22 shows an example of the instructor chat screen that appears when the internal chat icon is clicked. [Figure 23] Figure 23 shows an example of the admin center screen that appears when you click the admin center icon. [Modes for carrying out the invention]

[0024] One embodiment of the present invention will be described below with reference to the drawings.

[0025] <Summary of this embodiment> Figure 1 is an overall configuration diagram of a business support system according to an embodiment of the present invention.

[0026] Below, using Figure 1, we will explain the details of the learning management system for online students as a business support system.

[0027] As shown in Figure 1, the business support system of the present embodiment includes a plurality of user terminals 101, Internet / network 102, an information processing server (cloud) 103, a database (Supabase / PostgresSQL) 104, a storage (PDF / audio) 105, and an AI inference / generation module (LLM / GenAI) 106.

[0028] Although not shown in the drawings, the learning management system 1 for online students of the present embodiment includes the characteristic ToDo optimization management unit, automatic curriculum generation unit, and workflow-integrated AI generation unit of the present invention. These functions are implemented by software modules or the like stored on the information processing server 103. With respect to other functions, the portions included in known learning management functions in the information processing server 103 are omitted. Hereinafter, the characteristic ToDo optimization management unit, automatic curriculum generation unit, and workflow-integrated AI generation unit of the present invention will be described in detail.

[0029] <ToDo Optimization Management Unit (Daily ToDo Optimization and Management Function)> The ToDo optimization management unit has a daily ToDo optimization and management function.

[0030] <Background> In conventional courses and the like, it is common to manage learning progress on a monthly or weekly basis, and it is usual that online students need to make daily study plans and organize tasks by themselves. However, since poor self-management by online students may lead to causes of failure to obtain desired results, it is desirable to create and manage study plans for online students on a daily basis.

[0031] <Overview of ToDo Optimization Management Unit> Although not shown in the drawings, the ToDo optimization management unit includes a daily ToDo description step, a reminder message transmission step, and an automatic calculation of goal achievement curriculum and table of contents ToDo display reflection step.

[0032] <Daily To-Do List Notation Steps> The daily ToDo display step first involves the individual curriculum generated by the automatic curriculum generation unit being mapped to the daily schedule, and then displaying "What to do today" on the individual top page of each online student. In other words, the individual top page of each online student is displayed on the user terminal 101 via the internet / network 102 from the information processing server 103.

[0033] <Steps to send a reminder message> The reminder message sending step sends a reminder message to the online student's user terminal 101 via the official SNS (for example, LINE) if the daily ToDo items listed in the daily ToDo notation step above have not been completed by a time arbitrarily determined. In other words, the information processing server 103 sends the reminder message to the user terminal 101 via the internet / network 102.

[0034] <Automatic calculation of goal achievement curriculum and daily to-do display update steps> The goal achievement curriculum automatic calculation and daily to-do display reflection step defines the progress at 23:59 each day as "current progress," automatically recalculates the goal achievement curriculum for the following day and onward, and reflects it in the latest version of the daily to-do display using the same method as the daily to-do display step. In other words, the latest version of the daily to-do display is reflected from the information processing server 103 to the user terminal 101 via the internet / network 102.

[0035] The steps described above—the daily to-do list display step, the reminder message sending step, and the goal achievement curriculum automatic calculation / daily to-do display reflection step—represent only the minimum necessary processing; other processing may also be included.

[0036] <Automatic Curriculum Generation Unit (Automatic curriculum generation function to achieve goals)> The automated curriculum generation unit has the function of automatically generating a curriculum until the objective is achieved.

[0037] <Background> Traditional skill acquisition and learning courses primarily employ one of the following methods: (1) a uniform curriculum design for all students, (2) a curriculum design freely created by each student, or (3) a curriculum design individually assigned by the instructor to each student. None of these can be considered optimal for achieving individual goals. Ideally, to better support students in achieving their goals, it is desirable to create individualized curricula objectively and rationally, after thoroughly understanding each student's current situation, goals, and past experiences. Therefore, we developed this function with the belief that by collecting individual circumstances of students and using AI that has learned from the success stories of past students and monitor participants to design the most feasible curriculum for each student, we can increase the probability of students achieving their goals.

[0038] <Overview of the Automatic Curriculum Generation Unit> The automated curriculum generation unit, although not shown in the diagram, includes a dialogue and interview step with the student and a curriculum generation step for the student in question.

[0039] <Steps for conducting dialogue and interviews with participants> For online students, we will conduct a dialogue-based interview focusing on the following items to understand their current situation and goals. • Disposable time (weekdays and holidays) • Target results • Current skill level • Skills and work experience gained / learned in the past

[0040] <Curriculum generation steps for the student in question> Based on the learning progress and achievements database of online students collected by the ToDo Optimization Management Department, a curriculum for the current online student is generated based on the degree of match with past online student cases. In other words, the curriculum for the current online student is generated within the information processing server 103 based on the degree of match with past online student cases.

[0041] The above-mentioned steps for conducting dialogue interviews with students and generating curricula for those students represent the minimum necessary steps; other steps may also be included.

[0042] <Workflow-Integrated AI Generation Unit (Workflow-Integrated AI generation function optimized for generating learning deliverables)> The workflow-integrated AI generation unit has workflow-integrated AI generation functions optimized for generating learning deliverables.

[0043] <Background> In conventional skill learning support, a "learning-first" flow is common, where learners create deliverables through trial and error after learning. However, this method has the drawback of low learning efficiency and requiring a significant amount of time to reach practical-level quality because learning proceeds without understanding the "correct form." Ideally, to acquire practical skills in the shortest possible time, an "output-first" learning flow is desirable, starting with the structure and generation process of "high-quality deliverables required in the market," and learning is carried out in reverse from there. Furthermore, general-purpose generation AI lacks specific business know-how (prompt configuration and domain knowledge), making it difficult for beginners to obtain deliverables at a practical level. Therefore, we developed this function to maximize learning efficiency and ensure practical quality by learning and integrating the highly confidential knowledge of experts and optimizing the workflow itself on the system side.

[0044] <Overview of Workflow-Integrated AI Generation Unit> The automated curriculum generation unit, although not shown in the diagram, includes an optimization workflow template selection step, an AI dynamic hearing input completion step, and an optimization all-in-one AI generation step for generating learning deliverables.

[0045] <Optimization Workflow Template Selection Step> Based on web-completed measures (specific business domains) with high market demand, an "optimized workflow template" in which expert knowledge is structured in advance is selected.

[0046] <AI Dynamic Hearing Input Completion Step> Based on the selected workflow, AI performs dynamic hearing (input completion) for online students, and comprehensively extracts elements necessary for generating deliverables. At this time, a unique inference model (or a group of highly tuned prompts) that integrates the know-how of the lecturer and expert team is executed in the back end.

[0047] <All-in-one AI Generation Step for Optimization of Learning Deliverable Generation> Based on the extracted elements, everything from strategic design to specific productions (text, configuration proposals, measure plans, etc.) is generated in an all-in-one manner. Online students can advance learning while referring to the generated high-quality deliverables as "correct criteria", thereby connecting theory and practice at the same time and enabling skill acquisition in the shortest time.

[0048] The processing of the above-mentioned optimized workflow template selection step, AI dynamic hearing input completion step, and all-in-one AI generation step for optimization of learning deliverable generation describes minimal processing, and may include other processing.

[0049] Figure 2 is a functional block diagram of the information processing server 103 shown in Figure 1. As shown in Figure 2, the information processing server 103 includes a receiving unit 201, a context generation unit 202, an ambiguity calculation unit 203, a determination unit 204, a confirmation request output unit 205, a security unit 206, an Evidence Pack generation unit 207, a response generation unit 208, a log audit unit 209, an intent determination / tool ​​execution unit (Intent Router & Tool Calling) 210, a metadata pre-filter unit 211, a UIP area 301, a CIP area 302, an OIP area 303, a priority table 304, and a version history area 305. Here, the receiving unit 201, context generation unit 202, ambiguity calculation unit 203, determination unit 204, confirmation request output unit 205, security unit 206, Evidence Pack generation unit 207, response generation unit 208, log audit unit 209, intent determination / tool ​​execution unit 210, and metadata pre-filter unit 211 are located within the control unit. Additionally, the UIP area 301, CIP area 302, OIP area 303, priority table 304, and version history area 305 are located within the storage unit. The intent determination / tool ​​execution unit 210 analyzes the query and, if necessary, instructs direct SQL / API processing to the DB. The metadata pre-filter unit 211 narrows down the search target using tags (tag table) before performing vector searches, etc.

[0050] The information processing server 103 further comprises: a receiving unit 201 that receives queries input from a user terminal 101; an intent determination unit (intent determination / tool ​​execution unit 210) that interprets the intent of the query and determines whether the query is an inquiry involving statistical processing or calculation of specific data; a tool execution unit (intent determination / tool ​​execution unit 210) that, if the intent determination unit determines that the query is an inquiry involving calculation, bypasses the document text reading and semantic interpretation process by the generating AI and executes a direct search process or calculation process on an external database; and means for storing definitive numerical values ​​or extracted results obtained from processing on the external database as a set of premise facts (evidence pack) for answer generation. The generating AI is characterized in that it generates an answer in natural language based only on the definitive facts stored in the evidence pack.

[0051] The information processing server 103 further includes a tag management unit that assigns metadata (tags) linked to organization identifiers and location identifiers to student, instructor, teaching material, and comment data. When generating the evidence pack, it is characterized by physically pre-filtering the data to be searched using the metadata before performing a similarity search (vector search) by the generating AI, thereby eliminating the inclusion of irrelevant data.

[0052] <Overview of this business support system> The business support system according to the present invention is characterized by having an area for storing a "hierarchical interaction profile" in the memory of the information processing server 103. The hierarchical interaction profile has a three-tiered hierarchical profile structure: individual user, location, and organization. Furthermore, through ambiguity resolution logic, evidence packs, and shadow profile functions, a highly accurate and secure business support system is realized.

[0053] <Hierarchical Profile Structure> By maintaining terminology and rules defined at the user individual (UIP), location (CIP), and organization (OIP) levels, and overriding and integrating them in the order of "individual > location > organization," the system generates the most appropriate context for each situation.

[0054] <Ambiguous Resolution Logic> The generated context and query are compared, and if the ambiguity exceeds a threshold, the AI ​​will not automatically generate an answer but will instead be forced to perform a "clarification process" that always prompts the user for confirmation.

[0055] <Evidence Pack> The data that the AI ​​uses to generate answers will be limited to "intermediate data structures" that have been searched and extracted based on the user's permissions. This will prevent the AI ​​from physically accessing data outside of its permissions.

[0056] <Shadow Profile Function> It has a function to manage data of individuals (students, etc.) who have not yet logged into the system, and to seamlessly link this data when they create an account later.

[0057] <Effects of this business support system> This business support system will produce the following benefits: Firstly, AI will be able to accurately understand organization-specific terminology and context, significantly reducing misinterpretations of work instructions and irrelevant responses. Secondly, because the AI's reference data is limited to the "evidence pack," the risk of data from other organizations or locations being mixed into the responses can be structurally eliminated, even in a multi-tenant environment. Thirdly, the shadow profile function reduces the data registration burden during the initial system implementation phase, while enabling smooth user invitations.

[0058] <Details of the learning management system for online students> The following will describe in detail, with reference to the drawings, an example of the educational institution platform "YUKICHI," which is an embodiment of the present invention. Note that the present invention is not limited to educational institutions but is applicable to hierarchical organizations in general, such as companies and government agencies. A specific example of implementing the functions of the ToDo optimization management unit 11, the curriculum automatic generation unit 12, and the workflow integrated AI generation unit 13 shown in Figure 1B into this business support system will be described.

[0059] As shown in Figure 1A, the business support system for online student learning management consists of an information processing server 103 built on the cloud and multiple user terminals (teacher's PC, parent's smartphone, etc.) 101 connected via the internet 102. The information processing server 103 has a database 104 which includes an authentication database, a business database, and a vector database (for RAG). It also includes storage 105 for storing PDF files and audio files, and an AI inference / generation module 106 which has LLM or generative AI (GenAI).

[0060] Figure 3 is a conceptual diagram of the data structure and priority table for hierarchical profiles (UIP / CIP / OIP).

[0061] As shown in Figure 3, the hierarchical profile includes a UIP (User Profile) 401, a CIP (Context Profile) 402, an OIP (Organization Profile) 403, a priority table (priority order UIP > CIP > OIP) 404, version history information (v1, v2, v3...) 405, and an integration context 406. Here, the configuration of each UIP, CIP, and OIP profile is created with the same item names: intent default, confirmation threshold, term map, output style, and safety boundary. This enables unified integration.

[0062] Figure 4 is a flowchart of the process from query reception to evidence pack generation and response output. The processing steps in Figure 4 consist of query reception (step S1), query intent determination (does it involve calculation / statistics?) (step S1-A), (branch YES) numerical determination by direct DB execution (Tool Calling) (step S1-B) (proceed to S9 to store the numerical value), (branch NO) priority table reference (step S2), UIP>CIP>OIP integration → context generation (step S3), ambiguity score calculation (step S4), threshold determination (branch) (step S5, confirmation request output (YES type) (step S6), user response (return to S3) (step S7), permission filter (org / campus) (NO type) (step S8), pre-filtering of target data by metadata (tag) (step S8-B), Evidence Pack generation (step S9), response generation (only Evidence Pack generation is referenced) (step S10), and response output with source (step S11).

[0063] Figure 5 is a sequence diagram of the account integration process using shadow profiles. As shown in Figure 5, the UI (user interface) state transitions include the Join / QR screen 501, Login / Sign Up 502, Onboarding (Organization Creation / Joining) 503, Staff Dashboard (Chat + Selection Context) 504, Family Dashboard (Student / Parent) 505, and Settings (Team Management / Campus Management) 506. The following describes the processing operation of the business support system using Figures 1A and 2-5.

[0064] <Context generation using hierarchical profiling> The context generation unit 202 within the control unit performs the following processing upon receiving a query. (1) Load the "Company-wide prohibited terms (e.g., discriminatory language, unconfirmed medical diagnosis)" from the Organizational Information Platform (OIP). (2) Load the "location-specific abbreviation (e.g., 'PS1' = 1st school building)" from the location profile (CIP) and merge it into the OIP. (3) Load "individual output preferences (e.g., bullet points, conclusion-first presentation)" from the user profile (UIP) and merge them into the CIP.

[0065] At this time, based on the priority table 404, lower-level settings override higher-level settings, thereby generating consistent context information.

[0066] <Generation and Security of Evidence Pack> When the AI generates an answer, security is ensured through the following procedure. (1) Identify "accessible organization IDs and base IDs" from the user ID. (2) A security means (Security Definer function) extracts only data matching this ID from the database. (3) Add "source metadata (creation date and time, creator)" to the extracted data, and develop it as an "evidence pack" on the memory. (4) The answer generation unit (LLM) 208 receives only this evidence pack as input and generates an answer. At this time, the direct access path to the external DB is blocked.

[0067] <Deterministic Tool Calling to Prevent LLM Hallucination> In this system, the Generative AI (LLM) functions not only as a text generation tool but also as an "Intent Router." Typical LLMs have a tendency to skip lines or make calculation errors (hallucination) when performing statistical and mathematical calculations, such as "How many students scored 80 points or higher?", from large amounts of text data. To prevent this, when this system receives a user query, the LLM does not immediately read the document information but first classifies the intent of the query. If the query is determined to require a "database query (calculation, statistics, search)," it bypasses the LLM's text inference and directly triggers the query function (Tool Calling function) of a predefined API or database (such as Supabase). The database uses SQL, etc., to calculate the exact number (e.g., "14 students") and stores the determined absolute value in an "evidence pack." The LLM then uses the provided absolute value to generate the final natural language response (e.g., "There are 14 students who scored 80 points or higher."). This ensures 100% statistical accuracy, independent of the LLM's computational capabilities.

[0068] <Pre-filtering using universal tags (metadata)> Furthermore, to prevent hallucination and improve processing speed, this system performs pre-filtering using metadata (tags). Student characteristics, instructor expertise, teaching material types, and interview comments are all managed using universal tags based on the organization ID and location ID (e.g., #Math_Struggles, #LackOfConcentration). When the AI ​​performs information retrieval (vector search), it uses this metadata beforehand to physically narrow down (funnel) the data space to be searched. This structurally eliminates noise that could cause the LLM to mistakenly reference data from other subjects or unrelated students, building a highly accurate and concentrated "evidence pack."

[0069] <Shadow Profile and Invitation Flow> (1) The administrator registers the students' basic information (name, grade) in bulk using a CSV file or similar. At this point, since the students do not have login accounts, the data is stored in the "shadow profile area" and a unique identifier is assigned to each. (2) The system issues a QR code (registered trademark) with this identifier embedded in it. (3) When a parent or guardian scans the QR code (registered trademark) to register a new account, the account integration mechanism links the shadow profile with the new account and grants access to the stored data.

[0070] <Examples of applications (other than education)> This invention is also effective in general businesses. For example, by setting different definitions of the word "performance" in the branch profile (CIP) for the sales department (sales) and the development department (speed), it is possible to obtain the optimal answer for each department.

[0071] The following will provide a detailed explanation of web-based learning management, using "YUKICHI," an online learning management system developed by implementing the above-described invention, as an example.

[0072] Figure 6A shows an example of the initial web browser screen on user terminal 101 in the learning management system of this embodiment.

[0073] When a URL is entered in the browser window, the initial browser screen shown in Figure 6A is displayed. In this example, the web screen used by the administrators and instructors of an online tutoring school is displayed. At the top of the screen are tabs for selecting the school branch and student, and students whose learning management needs to be handled are displayed using the dropdown menus. At the bottom of the screen are functions such as an AI for performance analysis (e.g., "Analyze last math score"), an AI for summarizing (e.g., "Summarize teacher's comments"), and an AI for automatically creating emails (e.g., "Create an email for parents"). At the very bottom of the screen is a chat function for asking questions about students. Various operation icons are located on the right side of the screen. Specifically, there are icons for student information, calendar, pulse, comment, transcription, grades, career path, resources, Value, suggestion, internal chat, and management center. In Figure 6A, the internal chat and management center icons are hidden at the bottom, and they are displayed by scrolling up and down the scroll bar on the far right.

[0074] Figure 6B shows the icons in the lower right corner of the initial browser screen in Figure 6A, along with example comments summarizing the functions of each icon. Hovering the mouse cursor over each icon will display a description of its function. Specifically, the Student Information icon displays "a list of student basic information, alerts, and tasks," the Calendar icon "optimizes schedules by matching student skills and resources," the Pulse icon "assesss student well-being with AI tests and sentiment analysis," the Comments icon "a private timeline of teacher observations and student progress," the Transcription icon "records meetings, transcribes audio, and provides instant summaries," the Grades icon "tracks grades, visualizes progress, and uses AI to fill in missing data," the Career path icon "tracks changes in desired schools, future dreams, and goals in a timeline format," the Resources icon "tracks inventory of school supplies such as textbooks and iPads®," the Value icon "manages profit and loss for each student and provides AI-powered decision support," the Proposal icon "automatically generates professional-quality proposals and curricula for interviews and workshops based on all student data," the Internal Chat icon "a private communication tool between instructors and administrators," and the Administration Center icon "a dashboard for managing urgent alerts and important warnings."

[0075] Figure 7 shows an example of the student context screen that appears when the student information icon is clicked. The student context screen displays the student's name, the student's "Schedule and Course," "Pre-Class Briefing," and "To-Do / Messages from Management." In reality, the initial browser screen in Figure 6A has a fixed display area, so the lower area of ​​Figure 7 is displayed by sliding up and down the scroll bar (not shown) located on the right edge. The items displayed in "Schedule and Course" are the class the student is enrolled in, the days of the week they attend classes, the date and time of their next attendance, and the subjects they are enrolled in. "Pre-Class Briefing" displays birthday information, the scheduled start date of school periodic tests, the student's psychological state based on Pulse data, and appropriate responses. "To-Do / Messages from Management" displays documents to be collected from parents and their collection deadlines, as well as documents to be distributed to the student. Tasks can be added as needed.

[0076] Figure 8 shows an example of the schedule management screen that appears when the calendar icon is clicked. The calendar-style list displays the subject and instructor for each date, day of the week, and class period on a weekly basis, and is color-coded to allow identification even when the instructor is absent. Furthermore, by using the AI ​​concierge, it is possible to optimize the schedule through chat-style communication.

[0077] Figure 9 shows an example of the student wellness dashboard screen displayed when the pulse icon is clicked. The student wellness dashboard contains data entered by students based on their self-reported motivation, stress levels, sleep quality, etc. For example, a student's self-reported motivation of 80%, stress level of 45%, and sleep quality of 90% are quantified. These values ​​are compared and analyzed with the instructor's post-lesson student observation records, and the student's comprehension and concentration levels are displayed with a 5-star rating. Events that influenced changes in personality traits are automatically associated by AI from student comments and events in their personal lives. For example, related information such as "extroversion" and "conscientiousness" are assigned.

[0078] Figure 10 shows an example of the audio scriber screen that appears when the transcription icon is clicked. In the recording studio, you can record meetings or interviews, or upload existing audio files. Recording details include the recording title, attendees / speakers, and the context / instructions for the AI ​​summary. You can use the recording history to review past recordings, transcripts, and summaries.

[0079] Figure 11 shows an example of the student comment management screen that appears when the comment icon is clicked. The instructor view and administrator view can be displayed using the demo role switcher. YUKICHI Insight displays comments analyzed by AI, for example, "Students show fatigue patterns on Tuesdays due to club activities. Many of the math comments this week are related to 'concentration'." In addition, multiple instructors can view each other's observations and exchange comments.

[0080] Figure 12 shows an example of the grade management screen that appears when you click on a grade icon. The grade icons include report card, mock exam, regular test, and certification / other. In Figure 12, the report card grades (internal assessment scores) are displayed for each school year and semester.

[0081] Figure 13A shows an example of the career tracker screen (1) displayed when the career path icon is clicked, and Figure 13B shows an example of the career tracker screen (2) displayed when the career path icon is clicked. The lower area of ​​Figure 13B is displayed by sliding up and down the scroll bar (not shown) located on the right edge. The career tracker in Figures 13A and 13B allows you to track the trajectory of students' dreams.

[0082] The career tracker displays current career goals (first choice, future dreams, current objectives) and learning attitudes (favorite subjects, subjects that are difficult), and presents a timeline for achieving those goals. Records can be added freely, and the timeline can be modified at any time. Information on passing qualification exams can be added through self-reporting, such as passing the Eiken Grade 2 exam.

[0083] Figure 14 shows an example of the resource management screen that appears when a resource icon is clicked. The resource management screen displays an inventory list. For example, it includes items such as textbooks, iPads, copy paper, and summer course brochures. Automatic ordering is possible by setting a critical level in case of insufficient stock. The resource allocation matrix allows for tracking and managing resource distribution for each class or group.

[0084] Figure 15 shows an example of the resource import screen that appears when you click the "Upload Invoices / Inventory" button in Figure 14. By uploading invoices and delivery slips, the AI ​​can automatically analyze them and automate inventory management, and the AI ​​assistant can be used.

[0085] Figure 16 shows an example of the resource assignment screen that appears when you click the "Assign Resources" button in Figure 14. The resources are distributed to the target students or instructors, and the associated accounting processes are specified to assign the resources. This is executed by clicking the "Execute Assignment" button.

[0086] Figure 17A shows an example of the management dashboard screen (part 1) that appears when the Value icon is clicked. Here, a graph showing the profit trend (past 12 months) and revenue by school building are displayed.

[0087] Figure 17B shows an example of the management dashboard screen (part 2) that appears when the Value icon is clicked. In the course-specific profit center analysis, you can see how each course contributes to the overall profit, and AI analysis and suggestions are provided.

[0088] Figure 18 shows an example of the smart transaction registration screen that appears when you click the "Register Transaction" button in Figure 17A. On the smart transaction registration screen, you can register a transaction by entering the transaction details in natural language or by uploading a photo of the receipt.

[0089] Figure 19A shows an example of the suggestion generator screen that appears when the suggestion icon is clicked. The AI ​​deep research instruction combines the target student's data with external information to generate a detailed analysis report. The "Start Deep Research" button is displayed in the lower right corner of the screen.

[0090] Figure 19B shows an example of the AI-generated report display screen that appears when the "Start In-Depth Research" button in Figure 19A is clicked. The AI ​​generates an analysis report, and the curriculum analysis and current situation analysis of the target student's desired school are displayed as instructional suggestions.

[0091] Figure 20 shows an example of the display screen for the proposal slides for parents that appear when the "Convert to Slides" button in Figure 19B is clicked. The proposal slides for parents are proposal slides generated by AI based on the report, and they display a specific plan. The buttons for "Export to PDF" and "Generate Internal Script" are displayed in the lower right corner of the display screen.

[0092] Figure 21 shows an example of the display screen for internal materials (for department heads and instructors) that appear when the "Generate Internal Script" button is clicked. The internal materials are intended for department heads and instructors and display a report summary and a talk script / anticipated Q&A, showing a report and talk script to help conduct interviews successfully.

[0093] Figure 22 shows an example of the instructor chat screen that appears when the internal chat icon is clicked. If the AI ​​detects, for example, an instructor's absence from the message entered in the instructor chat, the following message will be displayed to the instructor: "The AI ​​has detected a possible absence. Would you like to notify the administrator and suggest detecting a substitute instructor?" The instructor can choose either <Detect Substitute> or <Ignore>.

[0094] Figure 23 shows an example of the admin center screen displayed when the admin center icon is clicked. Examples of urgent actions (Red Flags) include instructor absence notifications and inappropriate comments. Important warnings (Warnings) include a sharp drop in Pulse score and an increased risk of withdrawal, and the AI ​​can suggest action plans.

[0095] <Note> <Invention 1> A business support system comprising a user terminal connected via a network and an information processing server, wherein the information processing server comprises a storage unit that stores data associated with an organization identifier, a site identifier, and a user identifier, and a control unit that executes processing in response to input from the user terminal, the storage unit having a user profile area that stores vocabulary definitions and output preferences for each user, a site profile area that stores business rules and abbreviation definitions for each site, an organization profile area that stores control rules and common terminology definitions for each organization, and a priority table that defines the order of application when definition contents conflict between the user profile area, the site profile area, and the organization profile area, and the control unit comprising a receiving unit that receives queries input from the user terminal, a context generation unit that generates user-specific context information by overwriting or supplementing higher profile definitions with lower profile definitions according to the priority table, an ambiguity calculation unit that evaluates the consistency between the context information and the query and calculates an ambiguity score indicating semantic ambiguity, and an output unit that interrupts processing and outputs a confirmation request including choices to the user terminal when the ambiguity score exceeds a predetermined threshold.

[0096] <Invention 2> The business support system according to Invention 1 is characterized in that the control unit further comprises a security unit that manages access rights to the database, the security unit issues queries that limit the search range based on the user's organization identifier and location identifier, generates an intermediate data structure (evidence pack) containing only the acquired data set, the source metadata of the data, and the reference range identifier, and expands it into memory, and the control unit includes an answer generation unit that generates an answer by referring only to the information in the evidence pack while blocking direct access to external databases not included in the evidence pack.

[0097] <Invention 3> A business support system according to Invention 1 or 2, wherein the storage unit has a student database that stores at least quantitative indicators such as performance data, qualitative records such as interview record data, and psychological state data showing changes over time, linked to a student identifier, and the control unit has a proposal generation unit that refers to the student database and the evidence pack and generates proposal data including the current analysis results and future milestones.

[0098] <Invention 4> A business support system according to any one of Inventions 1 to 3, wherein the storage unit has a shadow profile area that holds attribute information of a person who does not have a login account, and the control unit comprises a detection unit that detects the reading of code information or access to an invitation link by the user terminal, a role determination unit that determines the user's authority role (administrator, instructor, or guardian) based on the type of code information, and an account integration unit that assigns the determined authority role to a newly created login account and links it with the data in the shadow profile area.

[0099] <Invention 5> A business support system according to any one of Inventions 1 to 4, characterized in that the control unit includes a profile update unit that updates the vocabulary definition of the user profile area based on the content of the answer only when it detects that a definitive answer has been input from the user terminal in response to a confirmation request from the output unit.

[0100] <Invention 6> The business support system described in Invention 2 is characterized in that the evidence pack is generated via a filtering layer that verifies whether it contains, in addition to the referenced data set, any "forbidden terms" or "forbidden categories" defined in the organizational profile area.

[0101] <Effects of Inventions 1-6> According to Inventions 1 to 6 described above, by prioritizing and integrating contexts across three levels—organization, location, and individual—to accurately identify user intent, and by confining AI-driven data access to a physically restricted area, it is possible to realize a highly accurate and secure business support system. [Explanation of Symbols]

[0102] 1. Learning management system for online students 101 User terminal 102 Internet / Network 103 Information Processing Server

Claims

1. In an online student learning management system comprising user terminals used by online students connected via a network and an information processing server, The aforementioned information processing server is The To-Do Optimization Management Department is responsible for optimizing and managing To-Dos on a daily basis, A curriculum automatic generation unit that automatically generates a curriculum to achieve the goal, A workflow-integrated AI generation unit that uses generation AI to integrate workflows optimized for generating learning deliverables, A context generation unit that generates user-specific context information by integrating three levels of profiles—organization, location, and individual user—in a predetermined order of priority, An evidence pack generation unit that physically pre-filters the search target data using metadata linked to the aforementioned organization and location, and generates an evidence pack that stores only the established facts extracted by the pre-filtering; It has, The aforementioned generating AI is characterized by generating responses in natural language based only on the established facts stored in the evidence pack, thereby providing a learning management system for online students.

2. The ToDo Optimization Management Unit includes a display step that incorporates the individual curriculum generated by the Automatic Curriculum Generation Unit into daily tasks and displays "What to do today" on the individual top page of the online student; a sending step that sends a reminder message to the online student's user terminal via the official SNS if the daily ToDo tasks displayed in the display step are not completed by a time arbitrarily set; and a reflection step that defines the progress status at 23:59 each day as "current progress," automatically recalculates the goal achievement curriculum for the following day and beyond, and reflects this in the display of the latest version of the daily ToDo using the same method as the display step. A learning management system for online students according to claim 1, characterized by comprising the above.

3. The automated curriculum generation unit includes a hearing step in which it conducts a dialogue-based interview with the online student to understand their current situation and goals, focusing on items such as available time (weekdays and holidays), target results, current skill level, and skills and work experience they have previously worked on and / or learned; and a curriculum generation step in which it generates a curriculum for the online student based on the degree of match with past online student cases, using the learning progress and achievement results database of the online student collected by the ToDo optimization management unit. The learning management system for online students according to claim 2, characterized by comprising the features described herein.

4. The workflow-integrated AI generation unit is characterized by comprising: a selection step of selecting a workflow template that defines a deliverable generation process, which is associated with a specific business domain and stored in advance in the storage unit of the information processing server, based on a selection input of a specific business domain entered from the user terminal; an extraction step in which the generation AI conducts dynamic interviews with the online student based on the workflow selected in the selection step and extracts elements necessary for the generation of deliverables; and an AI generation step of generating everything from strategic design to specific deliverables based on the elements extracted in the extraction step.

5. The learning management system for online students according to claim 4, characterized in that in the extraction step, a proprietary inference model or a highly tuned set of prompts that integrates the know-how of the instructor and expert team is executed in the backend.

6. The information processing server further comprises: a receiving unit that receives queries input from the user terminal; an intent determination unit that interprets the intent of the query and determines whether the query is a query involving statistical processing or calculation of specific data; a tool execution unit that, if the intent determination unit determines that the query is a query involving calculation, bypasses the document text reading and semantic interpretation process by the generating AI and performs a direct search or calculation process on an external database; and means for storing definitive numerical values ​​or extracted results obtained from processing by the external database as a set of premise facts (evidence pack) for answer generation, wherein the generating AI generates an answer in natural language based only on the definitive facts stored in the evidence pack, characterized in that the learning management system for online students according to claim 1.

7. The information processing server further comprises a tag management unit that assigns metadata (tags) associated with an organization identifier and a location identifier to student, instructor, teaching material, and comment data, and when generating the evidence pack, the system physically pre-filters the data to be searched using the metadata before performing a similarity search (vector search) by the generating AI, thereby eliminating the inclusion of irrelevant data, as described in claim 6.

Images