Adult learner-oriented AIGC application capability personalized training agent and implementation method

Abstract

The invention discloses an adult learner-oriented AIGC application capability personalized training agent and an implementation method thereof, an integrated architecture of an exclusive education knowledge base, a double-agent collaborative engine and a self-adaptive interaction engine is constructed based on a deduction platform OpenAPI, and a six-layer technical architecture is adopted: an infrastructure layer, a data layer, a service layer, an agent layer, an integration layer and an application layer. The method comprises the following steps: deploying double agents of a basic course (theoretical teaching) and a skill course (practical training evaluation), realizing accurate question and answer through a BERT-base vector library and'keyword + semantic similarity 'mixed retrieval, forming a learning closed loop in combination with a class test and a Web search plug-in, and generating a multi-device adaptive interface based on Vue3.0. According to the method, the knowledge retrieval response is smaller than or equal to 1.2 s, the semantic similarity is larger than or equal to 0.85, the learning efficiency is improved by 50% or above, the method has the public welfare attribute, and the economic and technical threshold of adult AIGC learning is broken through.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence education technology, and in particular relates to an intelligent agent and its implementation method for personalized training of AIGC application capabilities for adult learners. Background Technology

[0002] With the rapid development of AIGC technology, the demand for AIGC application skills among adults in society is growing. However, the existing AIGC education and training technologies are significantly lacking, mainly in three categories: (1) Standardized online courses: such as AIGC courses on the Coursera platform, which mainly consist of pre-recorded videos and fixed assignments, lack personalized interaction and scenario-based training, and cannot adapt to the learning characteristics of adults who have "large differences in basic knowledge and fragmented time"; (2) Single AI auxiliary tools: such as iFlytek Smart Learning Network, which only provides Q&A or test functions, and the search relies on keyword matching (without semantic understanding), resulting in fragmented functions and an inability to achieve the connection between "learning-practice-evaluation"; (3) General large models: such as ChatGPT, which supports open-ended questions and answers, but the knowledge is based on old training data (poor timeliness), lacks authority, and has no structured learning guidance and training evaluation mechanism.

[0003] None of the aforementioned technologies form a closed loop of "teaching, learning, assessment, and training," and most of them require payment, are not designed for adults, and cannot meet the personalized and low-cost AIGC learning needs of adults. There is an urgent need for a training program that takes into account systematicness, personalization, and public welfare attributes. Summary of the Invention

[0004] This invention provides a personalized training agent and implementation method for AIGC application capabilities for adult learners. The purpose is to respond to the national policy of promoting artificial intelligence education, and to provide a personalized training agent and implementation method for AIGC application capabilities for adult learners. This solves the problems of poor interaction, weak practical training, delayed feedback, outdated knowledge and high cost in existing technologies, lowers the learning threshold for adult AIGC, and promotes the improvement of AI literacy among the general public.

[0005] This invention employs a differentiated deployment of dual intelligent agents. The foundational component leverages a self-developed AI Lifelong Learning Knowledge Base and a BERT-based model to construct a vector library, combined with RAG hybrid retrieval to achieve accurate knowledge-based question answering. It also develops in-class quiz plugins and a web search plugin to complete a closed loop of knowledge point explanation, quizzes, and cutting-edge case studies. The skills component utilizes Workflow to design practical training processes, coupled with a large model evaluation interface to achieve prompt-based training and multi-dimensional assessment. Simultaneously, it generates adaptive interfaces and quick interaction entry points based on Vue. Ultimately, it achieves personalized learning guidance, real-time feedback, and scenario-based training, significantly improving learning efficiency compared to existing technologies. Furthermore, its public welfare nature breaks down economic barriers, meeting the regular and efficient AIGC learning needs of adults.

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0007] This invention relates to an intelligent agent and implementation method for personalized training of AIGC application capabilities for adult learners.

[0008] The following is a detailed explanation of the architecture layer by layer, combined with the architecture diagram:

[0009] (i) The infrastructure layer is the underlying support of the system, ensuring the resilience and stability of resources:

[0010] Cloud VMs: Utilize Linux CentOS cloud virtual machines to provide computing resources for various services and support dynamic scaling to cope with fluctuations in user volume;

[0011] Docker Compose & Packages: Encapsulates various service modules (such as front-end, back-end, and plugins) using containerization technology to achieve environment consistency and rapid deployment;

[0012] Storage: Includes a MySQL database (stores user data, learning behavior, and knowledge base metadata) and Minio object storage (stores course videos, knowledge base files, and user-uploaded training projects), ensuring data persistence and high availability.

[0013] (ii) The Data Layer is the "data hub" of the system, accumulating and transferring data across the entire data chain:

[0014] User Account: Stores user registration information (account, identity tags such as "zero experience" or "professional"), providing an identity basis for personalized services;

[0015] User Historical Data: Records user learning behavior (such as records of incorrect answers, mastery of knowledge points, and practical training projects), questions and feedback data, which are used for subsequent learning path optimization;

[0016] AI Education Resources: Stores AIGC educational resources, including knowledge base documents (Docs), teaching videos (Videos), and courseware (PPTs), which are the core materials for knowledge transfer;

[0017] User Survey: Stores user survey data (such as learning needs and feature satisfaction) for system iteration and optimization.

[0018] (iii) The Service Layer is the "execution center" of system functions, providing modular service support:

[0019] Application Web Service: Based on the Nginx+Vue3.0 technology stack, it realizes the rendering and interaction of the front-end interface, meets the requirements of adaptive interface, and ensures the layout adaptation between PC and mobile terminals.

[0020] Application Service: Using Node.js, Minio, and MySQL technologies, it handles business logic (such as user authentication, learning progress management, and agent invocation scheduling), and serves as the "scheduling center" for each module.

[0021] Plugin Service: Based on Python + Minio technology, it develops and manages custom plugins (such as in-class quiz plugins and web search plugins) to achieve differentiated functions such as "knowledge expansion" and "real-time assessment".

[0022] (iv) The Agent Layer is the "core carrier" of AIGC's educational functions, built on the CozePlatform:

[0023] Coze Platform: As the underlying foundation, it provides capabilities such as agent creation, API calls, plugin integration, and workflow orchestration, and is the ecological support for the entire technical solution;

[0024] RAG Knowledge Bases: AIGC-specific educational knowledge bases, containing structured knowledge of AIGC basics and applications. After processing by the BERT-base model embedding, they are formed into a vector library that supports semantic retrieval.

[0025] LLM: It adopts the deepseek-r1 large model as the "brain" of the intelligent agent, and is responsible for core logics such as knowledge understanding, natural language generation, and training evaluation.

[0026] Agents: Deploy two agent instances – “AIGC Fun Learning Tutor Basics” and “AIGC Fun Learning Tutor Skills” – to respectively carry out theoretical teaching and skills training functions, and implement differentiated services through the Coze API;

[0027] Plug-ins: Integrates four types of plugins—Flashcards (knowledge cards), Quiz (in-class quizzes), Websearch (web search), and Reporting (learning reports)—to expand the functional boundaries of the intelligent agent.

[0028] Workflows: Two types of workflows are designed—Classroom management (such as controlling the order of knowledge point explanation) and Prompt engineering (prompt engineering training process), and the training process is standardized through visual process arrangement.

[0029] (v) The Integration Layer is the "connection hub" of the system, enabling service integration across modules and platforms:

[0030] LLM Streaming Service: Based on Python + OAuth2.0 + SSE protocol + CozeAPI technology, it enables streaming responses for large models (such as real-time output of knowledge explanations) and ensures the real-time nature of interaction;

[0031] Agent Spacing Service: Based on OAuth 2.0 + CozeAPI technology, it enables the isolation and collaboration of multiple agents, ensuring that the functional boundaries of the basic and skill-based agents are clear and interoperable.

[0032] (vi) The Application Layer is the user's "interaction entry point," presenting the final service form:

[0033] Interactive Learning (AIGC Classes): This function provides knowledge explanations of intelligent agents in the basic section. Users can learn AIGC knowledge according to the logic of "basic concepts → core technologies → application scenarios".

[0034] Learning Suggestions (Resources): This corresponds to the knowledge expansion function of the Web search plugin and knowledge base, pushing cutting-edge cases and learning resource links to users;

[0035] Testing (Quiz|Test evaluation): This corresponds to the in-class quiz plugin, enabling real-time assessment of knowledge mastery and analysis of incorrect answers;

[0036] Training (Prompt engineering): This corresponds to the engineering training function of the intelligent agent in the skill section, allowing users to complete the training loop of "template reference → creation → evaluation → optimization".

[0037] AIGC Exclusive Educational Knowledge Base: Focusing on the learning needs of adults in the adult community, it constructs a two-tiered content system of "fixed core knowledge + dynamically expanded information."

[0038] Fixed core knowledge: The core content of the self-developed AIGC basics and applications course covers the basic concepts of AIGC (definition, differences from traditional AI), development path (from early generative models to the large model iteration process in 2024), core algorithms (the principles of Transformer architecture, diffusion model, etc.), and cutting-edge application scenarios (virtual teaching assistants in the education field, content generation in the marketing field, etc.). All of these are presented in a structured text format, suitable for adult learners with different backgrounds (zero-based focus on case explanations, and those with a foundation focus on technical details).

[0039] Dynamically expanded information: Real-time crawling of AIGC industry trends, innovative cases in the education field (such as the application of multimodal large models in vocational training), and updates to practical tools (such as the operation guide for lightweight AIGC generation tools) over the past 3 months through the web search plugin, with accessible links (such as industry reports and teaching videos) to ensure the timeliness of knowledge.

[0040] The present invention has the following advantages over the prior art:

[0041] (1) Significantly improved learning efficiency: Through the technical architecture of "AIGC exclusive educational knowledge base + dual intelligent agent collaborative engine + adaptive interaction engine", the knowledge retrieval response time is ≤1.2s and the semantic similarity is ≥0.85, ensuring efficient and accurate knowledge acquisition; the seamless connection between the dual intelligent agents from theoretical teaching to skills training, combined with real-time feedback from in-class quizzes and automatic jump to analysis of wrong questions, allows adult learners to quickly master AIGC knowledge and skills, improving learning efficiency by more than 50% compared to traditional recorded courses or single Q&A tools;

[0042] (2) The learning threshold has been greatly reduced: the adaptive interaction engine supports multiple devices (PC, mobile) and multiple input methods (text, voice) to adapt to different adult learning habits; the public welfare nature opens up all functions without economic threshold, breaking the barrier of "paying to unlock core content" of existing AIGC education tools, so that the majority of adults in society can easily access AIGC education resources;

[0043] (3) Functional completeness surpasses existing technologies: Most existing online education platforms are pre-recorded courses + fixed assignments, lacking personalized interaction and scenario-based training; although general large models support open-ended questions and answers, the knowledge authority and timeliness are insufficient and there is no structured learning guidance; while this invention constructs a complete closed loop of "knowledge-teaching-training", from theoretical explanation to skills practice and then to cutting-edge expansion, with comprehensive functions and natural connection between each link, solving the pain point of adult learners "cannot learn and cannot use";

[0044] (4) Personalized services outperform standardized solutions: In response to the characteristics of adults with “large differences in basic knowledge and diverse needs”, a dual-agent system of basic knowledge and skills is designed. Combined with the data layer to record and analyze users’ learning behavior, personalized teaching with “one path for each person” is achieved. Compared with the existing technology’s “one-size-fits-all” standardized service, it can better meet the differentiated learning needs of adults.

[0045] (5) Filling a Market Gap: Currently, domestic AIGC education projects for adults either focus on high-end paid training or only provide fragmented popular science content. There is a lack of integrated educational tools that deeply integrate advanced technologies (such as large models, plug-in development, and workflow orchestration) and have public welfare attributes. This invention is public welfare oriented and integrates multiple technology modules to create a full-process AIGC education intelligent body, filling the market gap for public welfare and systematic AIGC education tools for adults. It effectively promotes the improvement of AI literacy among the general public and has significant social benefits and market value.

[0046] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 An architectural diagram for inventing intelligent agents;

[0049] Figure 2 The basic section of this invention provides the user's initial interaction interface with the intelligent agent.

[0050] Figure 3 This is the interface for explaining "What is AIGC" in this embodiment of the invention;

[0051] Figure 4 This is the "evaluation-supplementation-re-practice" closed-loop interface in this embodiment of the invention;

[0052] Figure 5 This is the interface for explaining the core AIGC technology in this embodiment of the invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Disadvantages of existing technology:

[0055] Knowledge retrieval and response deficiencies: Existing technologies largely rely on keyword matching, lacking semantic understanding capabilities and failing to utilize a professional vector knowledge base. This results in low retrieval accuracy, slow response times, and frequent irrelevant answers. For example, when a user asks "What is AIGC?", only fragmented terminology explanations are returned, failing to accurately match core chapters. This invention constructs a vector library using the BERT-base model, combining "keyword + semantic similarity" hybrid retrieval to address the issues of insufficient knowledge retrieval accuracy and delayed response.

[0056] The learning loop is incomplete: Existing tools are fragmented, with Q&A, quizzes, and practical training disconnected, lacking automatic connection mechanisms and real-time feedback. For example, after Q&A, users need to manually switch platforms for practical training, and incorrect answers cannot be linked to the knowledge base for supplementary learning. This invention develops an in-class quiz plugin and practical training process to achieve a closed loop of knowledge point explanation - automatic question generation - real-time scoring - targeted supplementary learning, solving the problems of learning continuity and delayed feedback.

[0057] The content suffers from insufficient timeliness and authority: general-purpose large models rely on old training data and lack the ability to acquire cutting-edge information, making them prone to becoming outdated; online course content is fixed and difficult to update. This invention integrates a web search plugin to push cases from the past three months, combined with a self-developed knowledge base, to solve the problems of poor timeliness and lack of authority of AIGC knowledge.

[0058] Poor personalization and interactivity: Existing technologies lack differentiated function configurations, have weak interface adaptability, and lack quick interaction entry points. This invention solves the problems of lack of personalized services and inconvenient interaction by using differentiated label configurations for dual intelligent agents and an adaptive interface, coupled with shortcut buttons.

[0059] The purpose of this invention is to respond to the national policy on promoting artificial intelligence education, and to create an intelligent tool for AIGC education and training for adults. This tool aims to solve the problems of poor interactivity, weak practical training, delayed feedback, and insufficient timeliness of knowledge in existing AIGC adult education, thereby lowering the threshold for adults to learn AIGC and promoting the improvement of AI literacy among the general public.

[0060] This invention employs a differentiated deployment of dual intelligent agents. The foundational component leverages a self-developed AI Lifelong Learning Knowledge Base and a BERT-based model to construct a vector library, combined with RAG hybrid retrieval to achieve accurate knowledge-based question answering. It also develops in-class quiz plugins and a web search plugin to complete a closed loop of knowledge point explanation, quizzes, and cutting-edge case studies. The skills component utilizes Workflow to design practical training processes, coupled with a large model evaluation interface to achieve prompt-based training and multi-dimensional assessment. Simultaneously, it generates adaptive interfaces and quick interaction entry points based on Vue. Ultimately, it achieves personalized learning guidance, real-time feedback, and scenario-based training, significantly improving learning efficiency compared to existing technologies. Furthermore, its public welfare nature breaks down economic barriers, meeting the regular and efficient AIGC learning needs of adults.

[0061] Example 1: Teaching Process for the Basic Intelligent Agent "AIGC Definition"

[0062] (I) Preliminary Preparations

[0063] Knowledge Base Data: In the Basic Concepts section of the AIGC Foundation and Applications Knowledge Base, the "AIGC Definition" content is: "AIGC (Generative Artificial Intelligence) is a branch of artificial intelligence that uses algorithms and models to enable machines to autonomously generate various types of content such as text, images, and audio. Unlike traditional artificial intelligence, which focuses more on analysis and decision-making, the core of AIGC lies in 'creation'. Related examples include 'using ChatGPT to generate product marketing copy and using Midjourney to generate virtual scene images'." This content is stored in JSON format, with fields including Knowledge ID (K1001), category tag (basic concept), content text, related examples, and reference source ("Generative Artificial Intelligence White Paper 2024").

[0064] Agent Configuration: The AIGC Fun Learning Teacher Basic Edition agent (Agent ID: AGT-BASE-202507) adds the tags "AIGC Introduction" and "Knowledge Popularization" through the Kouzi platform / api / v1 / agent / config interface, and configures the knowledge explanation process as "receive questions → search the knowledge base → generate popular explanations → return to users".

[0065] (II) Teaching Process

[0066] User Question: The user enters "What is AIGC?" in the input box at the bottom of the adaptive interface and clicks the send button.

[0067] Knowledge retrieval: The system calls the Kouzi platform / api / v1 / agent / rag / hybridSearch interface. First, it matches the content with knowledge ID K1001 under the "Basic Concepts" category of the knowledge base using the keyword "What is AIGC". Then, it calculates the semantic similarity to 0.92 (≥0.85) and filters out the content.

[0068] Content Generation: In the basic section, the intelligent agent calls the deepseek-r1 model, combined with the related case of "using ChatGPT to generate product marketing copy and using Midjourney to generate virtual scene images," the explanation is: "AIGC is like an intelligent 'content creator.' For example, if you want to write marketing copy for a product, tell ChatGPT the product's features and target audience, and it can automatically generate the copy; if you want to design a virtual game scene, input the description into Midjourney, and it will get the corresponding image. It's different from previous artificial intelligence, which may have focused more on analyzing data, while AIGC directly creates new content."

[0069] Result returned: The explanation is returned to the user through an adaptive interface, with a response time of 1.1s (≤1.2s).

[0070] (III) In-class quizzes

[0071] Quiz Generation: After the knowledge explanation is completed, the system automatically calls the / api / v1 / plugin / quiz / generate interface to generate 3 multiple-choice questions:

[0072] Question 1: To which branch of artificial intelligence does AIGC belong?

[0073] Option A: Analytical Artificial Intelligence

[0074] Option B: Generative Artificial Intelligence

[0075] Option C: Decision-making artificial intelligence

[0076] Correct answer: B

[0077] Question 2: Which of the following is a typical application of AIGC?

[0078] Option A: Play music using a smart speaker

[0079] Option B: Generate marketing copy using ChatGPT

[0080] Option C: Monitor heart rate with a smartwatch

[0081] Correct answer: B

[0082] Question 3: The main difference between AIGC and traditional artificial intelligence lies in...

[0083] Option A: Do you need data?

[0084] Option B: Can they create content independently?

[0085] Option C: Whether or not it can connect to the internet

[0086] Correct answer: B

[0087] User Answering and Scoring: After the user completes the quiz, the system calls the / api / v1 / plugin / quiz / verify interface to score the user in real time. If the user answers question 3 incorrectly, the system automatically calls the knowledge base association interface and jumps to the "Differences between AIGC and Traditional Artificial Intelligence" section in the "Basic Concepts" chapter for further explanation.

[0088] Example 2: Training Process for "Designing Prompts for Technological Products" by Intelligent Agent in the Skills Section

[0089] (I) Preliminary Preparations

[0090] Training template: Configure the "Technology Product Promotional Prompt" training template as "[Product Type] + [Core Selling Points] + [Target Audience]" through the Kouzi platform / api / v1 / agent / skill / prompt / template interface.

[0091] Intelligent Agent Configuration: The AIGC Fun Learning Teacher Skills Edition Intelligent Agent (Agent ID: AGT-SKILL-202507) adds the tags "Prompt Project" and "Scenario Training" through the Kouzi platform / api / v1 / agent / config interface, and configures the training process as "Template Push → User Creation → Multi-dimensional Evaluation → Optimization Suggestions".

[0092] (II) Practical Training Process

[0093] Template push: After completing the "AIGC Application Area" learning in the basic section, the system prompts the user whether to enter the skills training. After the user confirms, they are redirected to the skills section interface. The skills section agent calls the / api / v1 / agent / skill / prompt / template interface to push the template "[Product Type] + [Core Selling Points] + [Target Audience]" to the user.

[0094] User-created: Users input custom prompts such as "Smart headphones + active noise cancellation + commuters".

[0095] Multi-dimensional evaluation: The system calls the button platform / api / v1 / llm / evaluate interface to evaluate from three dimensions: "accuracy", "innovation", and "completeness".

[0096] Accuracy: The product type is smart headphones, the core selling point is active noise cancellation, and the target audience is commuters. All three match, so it scores 9 points (out of 10, weight 30%, this item scores 9 × 0.3 = 2.7 points).

[0097] Innovation: Active noise cancellation is a common selling point for headphones, lacking differentiation, scoring 6 points (weight 30%, score for this item 6 × 0.3 = 1.8 points).

[0098] Completeness: Includes product type, core selling points, and target audience, earning 10 points (weight 40%, this item scores 10 × 0.4 = 4 points).

[0099] Total score: 2.7 + 1.8 + 4 = 8.5 points.

[0100] Optimization Suggestions: Based on the evaluation results, the system generated the following optimization suggestions: "Your prompts are excellent in terms of accuracy and completeness, but lack innovation. Active noise cancellation is a common feature of headphones. You could add some differentiating selling points, such as 'smart headphones + active noise cancellation + long battery life + suitable for commuters,' which would make the generated content more distinctive."

[0101] Example 3: Adaptive Interface for Multiple Devices

[0102] (a) Equipment type and interface adjustment

[0103] PC (Chrome browser):

[0104] Interface layout: It adopts a left-right layout, with the left side displaying the agent's name, function introduction, and quick interaction area, and the right side displaying the input area and content display area;

[0105] Fonts and Buttons: The font size for the top logo area is 16px, the font size for the middle introduction area is 14px, the button size for the quick interaction area is 120×40px, and the width of the text input box for the bottom input area is 400px.

[0106] Mobile devices (WeChat built-in browser):

[0107] Interface layout: It adopts a top-bottom layout, with the name and function introduction of the intelligent agent at the top, the quick interaction area in the middle, and the input area at the bottom;

[0108] Fonts and Buttons: The font size for the top logo area is 14px, the font size for the middle introduction area is 12px, the button size for the quick interaction area is 80×30px, and the width of the text input box for the bottom input area is 280px.

[0109] (ii) Interactive operation

[0110] Quick button interaction: When a user clicks the "What are the core technologies of AIGC?" button in the quick interaction area on the PC, the system responds within 0.4 seconds, calls the knowledge base retrieval interface, and returns an explanation: "AIGC's core technologies include the Transformer architecture, diffusion model, etc. The Transformer architecture is like an efficient information transmission network, enabling the model to better understand the context of text or images; the diffusion model generates images and other content by gradually adding and removing noise."

[0111] Voice input interaction: Users can tap the microphone icon in the bottom input area on their mobile devices and say "AIGC common tools". The voice recognition accuracy is 96% (≥95%). After the system recognizes the speech, it calls the knowledge base retrieval interface and returns the content "AIGC common tools include ChatGPT (text generation), Midjourney (image generation), Runway (video generation), etc."

[0112] like Figure 2 As shown, the initial user interface for the basic AIGC agent is positioned to guide adults to start learning AIGC basics; the content is marked as "AIGC Fun Learning Tutor (Basic Edition)," previewing the process of "4 knowledge points + case explanation + in-class quiz," and providing 5 preset question buttons (such as "What is AIGC") and a custom question box; the technical value is to reflect the quick entry design of the adaptive interaction engine, which is suitable for adults' fragmented learning.

[0113] The image shows an interactive interface titled "AIGC Fun Learning Tutor (Basic Edition)". The top of the interface displays the agent's identity description ("Explaining AIGC knowledge in a fun way"), the middle section previews the learning process ("4 knowledge points + case studies + quizzes"), and the bottom features 5 preset question buttons ("What is AIGC?", "What are the core technologies?", etc.) and a text input box.

[0114] Interface positioning: The first user interaction entry point for the basic AI agent, with the core function of guiding adult learners to start basic AIGC learning.

[0115] Technical value: It embodies the "shortcut entry design" of the adaptive interaction engine, which is suitable for adults' "fragmented learning and clear needs" habits, laying the foundation for subsequent personalized teaching.

[0116] like Figure 3 The above is a dedicated explanation interface for "What is AIGC";

[0117] Positioning: To popularize the core concepts of AIGC; Content: To explain AIGC using the metaphor of a "knowledgeable little genius," to illustrate the role of prompts ("creative instructions"), to include a video link to the "prompt word project," and to generate 3 multiple-choice questions and the "continue learning" option; Technical Value: To verify the accuracy of RAG hybrid retrieval and its ability to automatically generate quizzes.

[0118] The image shows an interface titled "'What is AIGC' Explanation Interface". The top of the interface explains the concept of AIGC (including the metaphor of "Little Genius"), the middle section explains the function of the prompts and provides a video link, and the bottom section contains three multiple-choice questions (options A / B / C) and a "Continue Learning" input prompt.

[0119] Technical Value: This not only demonstrates the accuracy of RAG hybrid retrieval—the system matches the "Basic Concepts" chapter of the knowledge base with the question "What is AIGC?"—but also showcases the "automatic question generation" capability of the in-class quiz plugin. The questions closely adhere to the core knowledge of the "AIGC Basic Concepts" chapter in the knowledge base, and the generation logic meets the design requirements of "full coverage of knowledge points and reasonable distraction in the options." Simultaneously, it verifies the synergy between the large model and the knowledge base. The questions are generated by the large model in conjunction with the content of the "Basic Concepts" module of the knowledge base, ensuring the accuracy and relevance of knowledge delivery.

[0120] like Figure 4 The image shows the closed-loop interface for "evaluation-supplementation-repair";

[0121] Positioning: To provide feedback and expansion through test answering; Content: Question-by-question scoring and analysis (incorrect questions are reviewed by accessing the knowledge base), and 3 new questions are pushed based on mastery (extending to the "Technology Application" knowledge point); Technological value: Demonstrating the completeness of the learning loop and dynamic path planning ability.

[0122] The image is titled "Test Scoring and Remedial Learning Interface". The top of the interface shows the user's answer results ("Question 1 A is correct, Question 3 C is incorrect"), the middle section shows the analysis of the incorrect questions (referencing knowledge base content), and the bottom section shows 3 new multiple-choice questions.

[0123] Technical Value: First, it validated the completeness of the "learning closed loop" technology. Scoring was completed through the "verify interface" of the quiz plugin, followed by parsing generated by the large model in conjunction with the knowledge base. Finally, it triggered the in-class quiz plugin to generate new questions, achieving an automated flow of "assessment-supplementation-practice." It also demonstrated the "dynamic path planning" capability of the adaptive learning engine. Based on user answer data (types of incorrect answers, level of knowledge mastery), the system automatically adjusted subsequent learning content (extending from basic concepts to technical applications), aligning with the intelligent learning service goal of "personalized paths for each individual."

[0124] like Figure 5 The image shown is the interface for explaining AIGC's core technologies.

[0125] Positioning: Explaining AIGC core technologies and guiding continuous learning; Content: Using the metaphor of "building a castle" to explain technologies such as machine learning and deep learning, with links to technical videos, generating 3 multiple-choice questions and an "continue learning" entry; Technical value: Popularizing abstract technologies to ensure a coherent learning process.

[0126] The image is titled "AIGC Core Technology Explanation Interface". The upper part of the interface uses the metaphor of "building a castle" to introduce the topic. The middle part contains explanations of technologies such as machine learning and deep learning, as well as video links. The lower part contains three multiple-choice questions and a "Continue Learning" button.

[0127] Technical Value: By combining metaphors and case studies, the abstract core technologies of AIGC are explained in a simple and accessible way, lowering the learning threshold and helping learners understand and master them. The "Continue Learning" function reflects the coherent design of the learning process. Relying on the system's interactive and content push technology, it can automatically or manually guide learners to subsequent related knowledge learning according to their learning progress, ensuring the smoothness and continuity of the learning process and improving the learning experience and efficiency.

[0128] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A personalized training agent for AIGC application skills for adult learners, characterized in that: Built upon the OpenAPI architecture of the button platform, it includes a six-layer technical architecture and a dual-agent collaborative module. The six-layer technical architecture, from bottom to top, is as follows: (1) Infrastructure layer: Linux CentOS cloud virtual machine is used to provide computing resources, Docker Compose is used to encapsulate service modules, and MySQL database and Minio object storage are configured to achieve data persistence; (2) Data layer: Stores user account information, learning history data, AIGC educational resources and user survey data; user account information includes identity tags such as "zero-based" and "professional"; learning history data includes records of wrong questions and practical training works; AIGC educational resources include knowledge base documents and teaching videos; (3) Service layer: Web client interaction is implemented based on Nginx+Vue3.0, business logic is processed through NodeJs+Minio+MySQL, and plugin services are developed based on Python+Minio; (4) Intelligent Agent Layer: Deploy the dual intelligent agents "AIGC Fun Learning Tutor Basic Edition" and "AIGC Fun Learning Tutor Skill Edition", integrating RAG knowledge base, deepseek-r1 large model, in-class quizzes, Web search, knowledge flashcards, learning report plugins and classroom management, prompt training workflow; RAG knowledge base is a vector library built by BERT-base model; (5) Integration layer: Implement large-scale model streaming services based on Python + OAuth 2.0 + SSE protocol, and realize dual-agent isolation and collaboration through OAuth 2.0 + CozeAPI; (6) Application layer: Provides AIGC interactive courses, learning resource push, in-class quizzes, prompts and engineering training functions; the dual-agent collaborative module realizes accurate knowledge question answering through "keyword + semantic similarity" hybrid retrieval, and forms a learning closed loop of "knowledge point explanation → automatic question generation → real-time scoring → targeted supplementary learning → skills training" by combining in-class quiz plugin and training process, and all functions are open to users for free.

2. The AIGC application capability personalized training intelligent agent for adult learners according to claim 1, characterized in that, The Linux CentOS cloud virtual machine in the infrastructure layer supports dynamic expansion. The service modules encapsulated by Docker Compose include front-end interface modules, back-end business modules, and plugin modules, ensuring environment consistency and rapid deployment.

3. The AIGC application capability personalized training intelligent agent for adult learners according to claim 1, characterized in that, The AIGC educational resources in the data layer include "fixed core knowledge + dynamic extended information"; The core knowledge is the self-developed AIGC basics and applications course content, including basic concepts, core algorithms, and application scenarios; the dynamic expansion information is the AIGC industry case studies and tool updates crawled by the web search plugin in the past 3 months.

4. The AIGC application capability personalized training intelligent agent for adult learners according to claim 1, characterized in that, The RAG knowledge base of the agent layer is processed by the BERT-base model embedding to form a vector library. During hybrid retrieval, the knowledge base is first classified by keyword matching, and then the semantic similarity is calculated to filter the target knowledge. The threshold of the semantic similarity is ≥0.

85.

5. The AIGC application capability personalized training intelligent agent for adult learners according to claim 1, characterized in that, The application layer's prompts and engineering training functions include training template push, user creation reception, multi-dimensional evaluation, and optimization suggestion generation; the training template copy format is: [Product Type] + [Core Selling Points] + [Target Audience], and the multi-dimensional evaluation is based on accuracy, innovation, and completeness, with weights of 30%, 30%, and 40%, respectively.

6. The AIGC application capability personalized training intelligent agent for adult learners according to claim 1, characterized in that, The adaptive interaction engine supports both PC and mobile devices, and also supports voice input.

7. A method for implementing personalized training of AIGC application capabilities for adult learners, based on the AIGC application capability personalized training intelligent agent for adult learners as described in any one of claims 1-6, characterized in that, Includes the following steps: S1: Users register and label their identity through an adaptive interface, and the system stores user information in the data layer; S2: Basic Chapter The agent receives user questions, calls the hybrid retrieval interface to match RAG knowledge base content, generates popular explanations through the deepseek-r1 large model, and automatically triggers the in-class quiz plugin to generate 3 multiple-choice questions; S3: After the user completes the test, the system scores the test in real time, and incorrect questions are automatically redirected to the corresponding chapter in the knowledge base for analysis; S4: After completing the basic learning, the system prompts you to enter the skills training. The skills agent pushes the training template, receives user-defined prompts, conducts multi-dimensional evaluation, and generates optimization suggestions. S5: The system pushes AIGC cases from the past 3 months through a web search plugin and optimizes subsequent learning paths based on user learning data.

8. The method for personalized training of AIGC application capabilities for adult learners according to claim 7, characterized in that, The specific process of hybrid retrieval in step S2 is as follows: call the Kouzi platform / api / v1 / agent / rag / hybridSearch interface, first match the knowledge base category tags by keywords, then calculate the semantic similarity between the question and the knowledge base content, and filter knowledge items with a similarity ≥ 0.

85.

9. The method for personalized training of AIGC application capabilities for adult learners according to claim 7, characterized in that, The specific calculation method for the multi-dimensional evaluation in step S4 is as follows: Total score = Accuracy score × 30% + Innovation score × 30% + Completeness score × 40%, where the maximum score for each dimension is 10 points.

10. The method for personalized training of AIGC application capabilities for adult learners according to claim 7, characterized in that, The learning path optimization described in step S5 is based on user historical data. Basic concept courses are prioritized for "zero-based" users, and industry scenario training is prioritized for "professionals". The historical data includes the types of questions answered incorrectly, training scores, and learning time.

Images