Method for realizing multi-agent blueprint arrangement by zero code in AIGC platform education scene

Abstract

The invention discloses a method for realizing multi-agent blueprint arrangement by zero codes in an AIGC platform education scene, which comprises the following steps: step 1, designing a graphical blueprint editing interface, realizing connection arrangement processing of an AI basic large model and various different agent intelligent agent education module components, and enabling a user to edit a blueprint by dragging and connecting operations on the blueprint editing interface to obtain a blueprint editing interface; and forming a graph link. According to the method, the use threshold of the AIGC platform in an education scene is remarkably reduced, the requirements of complex code writing and professional technical knowledge are avoided through a visual graphical interface and zero-code operation, a large number of education workers can easily apply the multi-agent technology, the audience range of the multi-agent technology in the education field is greatly expanded through the arrangement, and the education experience is improved. More users can use the agent technology, and educational fairness and popularization are promoted.

Description

Technical Field

[0001] The present invention relates to the field of intelligent education technology, and specifically to a method for implementing blueprint orchestration of multiple agents with zero code in an AIGC platform education scenario. Background Art

[0002] AIGC (Artificial Intelligence Generated Content) refers to the process of automatically generating multimedia content such as text, images, audio, and video using artificial intelligence technology. In recent years, with the rapid development of deep learning and natural language processing technologies, AIGC has shown great potential and value in many fields.

[0003] Based on some content characteristics of AIGC, the application of AIGC platform in the field of education is gradually becoming a trend, bringing many new possibilities to teaching activities.

[0004] In the existing AIGC platform, there are a series of significant problems when it comes to the processing of multi-agent collaboration. First, the complex code writing requirements have become an insurmountable obstacle. Whether it is educators or ordinary users, in order to achieve effective collaboration between multiple agents, they often need to have deep programming skills and professional technical knowledge. This not only increases the difficulty of use, but also consumes a lot of time and energy. In addition, the existing technology requires users to have complex code writing capabilities and professional technical knowledge, which has formed a huge obstacle for educators and ordinary users.

[0005] Moreover, judging from the prompt words and the various functional characteristics of large models, the learning cost is too high for ordinary people. There are many different types of large models flooding the market, and users are often confused when choosing. They also lack a clear understanding of the applicable scenarios of different models.

[0006] Traditional orchestration methods perform poorly in terms of efficiency. Their rigid processes and complex operations make it difficult to meet the actual needs of fast and flexible customization in educational scenarios. Educational scenarios usually require rapid adjustment and optimization of multi-agent collaboration methods based on student characteristics, changes in teaching content, and innovations in teaching methods. However, traditional orchestration methods cannot respond to these changes in a timely manner and are difficult to adapt to the needs of rapid changes and personalized customization in educational scenarios. To this end, we propose a blueprint orchestration implementation method for zero-code multi-agent in educational scenarios on the AIGC platform. Summary of the invention

[0007] The purpose of the present invention is to provide a method for implementing blueprint orchestration of multiple agents in an AIGC platform education scenario with zero code, so as to solve the problems that need to be solved in the background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solution: a method for implementing blueprint arrangement of multiple agents with zero code in an AIGC platform education scenario, wherein the method for implementing blueprint arrangement of multiple agents with zero code in an AIGC platform education scenario comprises the following steps: Step 1: Design a graphical blueprint editing interface to realize the connection and arrangement between the AI ​​basic big model and various agent intelligent body education module components. Users can drag and connect in the blueprint editing interface to form a graph link, output results through the AI ​​big model or execute agent intelligent body education module components; Step 2: predefine the agent module and divide the agent module into multiple sub-agent intelligent education modules according to the multi-faceted definition of the education scenario; Step 3: Define the attributes and behavior rules of the sub-agent intelligent body education module, customize the attributes and behavior rules of the sub-agent intelligent body education module according to the personalized configuration, and the user can realize personalized agent configuration through parameter setting and condition selection; Step 4: Automatically generate execution logic and run to achieve results. Automatically generate the underlying execution logic. According to the line planning connected to the graphical blueprint editing interface and the sub-agent intelligent body education module on the line, rely on the blueprint editor to generate a directed graph, identify the nodes and edge nodes on the directed acyclic graph line in the directed graph, and form text logic with graph links. The generated text logic is fed to the AI ​​big model, and the AI ​​big model's own intelligent body is used to perform actions and observe environmental reactions, and adjust its subsequent behavior based on these observations, and repeat ability identification. Interact through the ReAct library, and finally the AI ​​big model returns the result; Step 5: Generate a general protocol interface that can access the graph API based on the orchestrated agent capabilities.

[0009] Traditional orchestration methods perform poorly in terms of efficiency. Their rigid processes and complex operations make it difficult to meet the actual needs of fast and flexible customization in educational scenarios. Educational scenarios usually require rapid adjustment and optimization of multi-agent collaboration methods based on student characteristics, changes in teaching content, and innovations in teaching methods. However, traditional orchestration methods cannot respond to these changes in a timely manner and are difficult to adapt to the needs of rapid changes and personalized customization in educational scenarios. The present invention significantly lowers the threshold for using the AIGC platform in educational scenarios. Through an intuitive graphical interface and zero-code operation, it avoids the need for complex code writing and professional technical knowledge, allowing educators, regardless of their technical background, to easily use multi-agent technology. This setting greatly expands the audience range of multi-agent technology in the field of education, allowing more users to use agent technology and promote educational equity and popularization. In addition, the present invention greatly improves the efficiency and flexibility of multi-agent arrangement. Users can quickly complete the arrangement and configuration of multi-agents through simple operations such as dragging, connecting, and setting parameters. At the same time, this flexible arrangement method enables the system to quickly respond to various changes in demand that frequently occur in education and teaching. Whether it is the adjustment of course content, the innovation of teaching methods, or the satisfaction of students' personalized needs, they can be reflected and supported in the multi-agent system in a timely manner. As a further description of the above technical solution: In the step 1, the graph link freely defines loop edges and loop nodes in the graph structure, and finally forms a directed acyclic graph. Through the intention recognition capability of the AI ​​basic big model, the connection line is automatically selected, and the agent intelligent body education module components on the execution line are selected. According to different user scenarios, the required big model and different agent intelligent body education module components are selected, and the relationship and task flow between the agent intelligent body modules are defined. The user starts the execution of the graph link through a Request request.

[0010] As a further description of the above technical solution: The educational scenario in step 2 includes teacher teaching, student learning, teacher and student evaluation, school administration, and school safety, and defines multiple aspects to divide the agent module into multiple sub-agent intelligent education modules.

[0011] As a further description of the above technical solution: The sub-agent intelligent body education module includes but is not limited to a teaching auxiliary agent, a question-answering agent, and an evaluation agent. The sub-agent intelligent body education module is classified according to definitions, and different permission access mechanisms are set when setting the classifications. Different sub-agent intelligent body education modules are customized and added according to different permission access mechanisms as needed.

[0012] As a further description of the above technical solution: The sub-agent intelligent body education module in the step three is used for agent execution operations, and the agent execution operations are composed of API communication protocols or executable codes. The user sets different parameters and different condition weights as the basis for judgment through the AI ​​basic model.

[0013] As a further description of the above technical solution: In the process of generating the API communication protocol in step five, a unique UUID random string is formed by combining the generated orchestrated graph link, and the effect is achieved through execution logic and operation to realize human-computer interaction. Through the unified API http request protocol, different orchestrated graph links are identified, and then this interaction is initiated. Respond returns the http result, and the selected graph connection is defined in the http parameter to form a proprietary graph API communication protocol interface.

[0014] As a further description of the above technical solution: The repeated ReAct in step 4 includes the following steps: Initialization: The sub-agent education module starts with a certain initial strategy or a completely random behavior pattern; Action: The sub-agent education module selects an action based on the current policy and applies it to the environment; Observation: The environment responds to the actions of the sub-agent education module and produces new states or results. The new state or result information is collected by the sub-agent education module as observations. Evaluation and learning: Based on the results collected from observations, the sub-agent education module evaluates the effectiveness of its actions and updates its own strategy based on the learning algorithm; Iteration: The above steps are performed repeatedly. After each cycle, the sub-agent education module may improve its behavior strategy until the preset learning goal or performance standard is reached.

[0015] As a further description of the above technical solution: The learning algorithm in the iterative ReAct adopts the policy gradient method, and the calculation formula of the policy gradient method is: ; in, ∇θ is the strategy parameter; J( θ ) is the objective function, usually expressed as the expected return under the strategy; Eτ~πθ is the expectation operator, the expected value of all possible trajectories τ, where τ is sampled according to the strategy πθ; τ is a complete trajectory, that is τ =(s0,a0,r1,s1,a1,r2,...,sT); πθ(at|st) is the probability of selecting action at according to the current strategy πθ in state st; logπθ(at|st) is the log value of the probability of the strategy choosing an action; ∇θlogπθ(at|st) is the gradient of the log value of the probability of the policy selecting an action with respect to the parameter θ; Rt is The cumulative return starting from time step t; in, ; in, R is the cumulative reward starting from time step t, which represents the sum of all future rewards from the current time step t to the terminal time step T; It represents the accumulation from k=0 to k=Tt, where T is the total number of time steps of the trajectory and t is the current time step. Therefore, Tt represents how many time steps are left from the current time step to the end of the trajectory. γ is the discount factor, and its value range is 0<γ≤1; γk is the discount for future rewards; k is the index of the time step, starting from 0; is the immediate reward obtained at time step t+k+1.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. Lower the threshold for use and popularize multi-agent technology; The present invention significantly lowers the threshold for using the AIGC platform in educational scenarios. Through an intuitive graphical interface and zero-code operation, it avoids the need for complex code writing and professional technical knowledge, allowing educators, regardless of their technical background, to easily use multi-agent technology. This setting greatly expands the audience range of multi-agent technology in the field of education, allowing more users to use agent technology and promoting educational equity and popularization. 2. Improve scheduling efficiency and flexibly respond to needs; The present invention greatly improves the efficiency and flexibility of multi-agent arrangement. Users can quickly complete the arrangement and configuration of multi-agents through simple operations such as dragging, connecting and parameter setting. At the same time, this flexible arrangement method enables the system to quickly respond to various demand changes that frequently occur in education and teaching. Whether it is the adjustment of course content, the innovation of teaching methods, or the satisfaction of students' personalized needs, they can all be reflected and supported in the multi-agent system in a timely manner. 3. Promote technology application and innovate education models; In addition, the present invention has effectively promoted the widespread application of AIGC technology in the field of education, provided educators with more powerful tools and means, and encouraged them to explore and update new educational models. With the help of multi-agent technology, it can realize personalized learning, intelligent tutoring, automated evaluation and other innovative educational models, improve the quality of education, and cultivate students' innovative thinking and autonomous learning ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the system principle of the present invention; Figure 2 This is a schematic diagram of the execution logic system of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Embodiment 1:

[0019] See also Figure 1-Figure 2 The present invention provides a technical solution: a method for implementing blueprint arrangement of multiple agents with zero code in an AIGC platform education scenario. The method for implementing blueprint arrangement of multiple agents with zero code in an AIGC platform education scenario comprises the following steps: Step 1: Design a graphical blueprint editing interface to realize the connection and arrangement between the AI ​​basic big model and various agent intelligent body education module components. Users can drag and connect in the blueprint editing interface to form a graph link, output results through the AI ​​big model or execute agent intelligent body education module components; Among them, through the graphical blueprint editing interface, users can implement complex logical processes without writing a lot of code, which greatly reduces the development threshold. The drag-and-drop interactive method makes it easy for non-technical personnel to get started, reducing the learning cost. The graphical blueprint editing interface allows users to directly see the relationship between each component and the data flow path, which is easy to understand and debug. In addition, users can freely combine different agent intelligent body education modules according to their needs and explore new application scenarios. When faced with changes in requirements, they can adapt to new requirements by simply adjusting the blueprint instead of rewriting the code. And with the powerful processing power of AI big models, it can automatically analyze and optimize bottlenecks in the entire workflow, improve overall efficiency, and provide users with smarter decision-making recommendations by integrating advanced algorithms such as reinforcement learning and deep learning.

[0020] Step 2: predefine the agent module, and divide the agent module into multiple sub-agent intelligent education modules according to the multi-faceted definition of the education scenario. The education scenario includes the teacher's teaching, the student's learning, the teacher's and the student's evaluation, the school's administrative management, and the school's safety. Each sub-agent intelligent body education module focuses on a specific functional area and can be independently developed, tested and updated, reducing interdependence and overall complexity. When adding or improving a sub-agent in a certain aspect, only the corresponding sub-agent intelligent body education module needs to be modified without affecting other parts. Moreover, users can select and configure different sub-agent intelligent education modules according to actual needs, build personalized education solutions, and optimize each sub-agent intelligent education module according to the characteristics of different education scenarios to improve performance and service quality; And through the graphical blueprint editing interface, users can intuitively understand and operate each sub-agent intelligent body education module, which lowers the threshold for use. Based on the segmented functional modules, it can provide a more refined service experience that fits user needs. Moreover, different sub-agent intelligent education modules can collect data in their respective fields, comprehensively covering all aspects of the education process, and use AI big models to integrate data from various sub-agent intelligent education modules to conduct deep learning and pattern recognition, providing a scientific basis for educational decision-making.

[0021] Finally, according to the functional characteristics of different sub-agent intelligent education modules, corresponding access rights are set to ensure data security. Differentiated privacy protection measures are also formulated for different types of data and application scenarios. When the needs of a certain aspect change, the relevant sub-agent intelligent education modules can be adjusted to respond quickly to maintain the vitality of the system. In addition, different sub-agent intelligent education modules can communicate and collaborate with each other to jointly complete complex tasks, such as personalized learning path planning. In summary, by subdividing the agent module into multiple sub-agent education modules, higher flexibility, customization, and intelligence can be achieved in the application of educational technology, thereby providing better services and support for educators and students.

[0022] Step 3: define the attributes and behavior rules of the sub-agent intelligent body education module. In order to overcome the problem of lack of personalized configuration in the prior art, the attributes and behavior rules of the sub-agent intelligent body education module are customized according to the personalized configuration. The user implements personalized agent configuration through parameter setting and condition selection. The sub-agent intelligent body education module in step 3 is used for agent execution operation. The agent execution operation consists of API communication protocol or executable code. The user sets different parameters and different condition weights as the basis for judgment through the AI ​​basic large model. Among them, the main core of the sub-agent module is the agent performing a certain operation, such as the student growth assessment agent in agent teaching. The key entity capabilities of its core student assessment can be composed of API communication protocols or executable codes. At this time, different conditional parameters need to be set as the basis for judgment, which may include key parameter information such as "grades", "extracurricular activities", "interests and hobbies", "daily consumption", and "daily activities". Conditional rules are set, such as "priority judgment", "extraordinary conditions", "low priority", "high priority", "loop judgment", "greater than", "less than" and other key conditions. According to the AI ​​evaluation robot that the user wants to realize, different parameters and different condition weights are set, so that the multi-agent system can better adapt to the specific educational environment and user needs, and provide more accurate and more practical services.

[0023] Step 4: Automatically generate execution logic and run to achieve results. Automatically generate the underlying execution logic. According to the line planning connected to the graphical blueprint editing interface and the sub-agent intelligent body education module on the line, rely on the blueprint editor to generate a directed graph, identify the nodes and edge nodes on the directed acyclic graph line in the directed graph, and form text logic with graph links. The generated text logic is fed to the AI ​​big model, and the AI ​​big model's own intelligent body is used to perform actions and observe environmental reactions, and adjust its subsequent behavior based on these observations, and repeat ability identification. Interact through the ReAct library, and finally the AI ​​big model returns the result; Step 5: Generate a general protocol interface that can access the graph API based on the orchestrated agent capabilities. Embodiment 2:

[0024] In the process of generating the API communication protocol in step five, a unique UUID random string is formed by combining the generated orchestrated graph link, and the effect is achieved through execution logic and operation to realize human-computer interaction. Through the unified API http request protocol, different orchestrated graph links are identified, and then this interaction is initiated. Respond returns the http result, and the selected graph connection is defined in the http parameter to form a proprietary graph API communication protocol interface.

[0025] When the graph API general protocol interface is generated, the following is a typical generated interface and request format: Request interface curl-XWS 'wss: / / HOST:PORT / api / v1 / third_interface / assistant / chat?chai_id=Chat ID&key=Secret key&token=Token / access_token' Request Format {"chatHistory":[{"is_bot":false,"source":0,"message":{"input":"Which courses did Li Ming study today","file_list":[]},"extra":"\"{}\"","type":"human","category":"question","flow_id":"ac5da15cc68f4b289a1123b2486c551f","chat_id":"5f85f74988e830aad4332b09d7f13eab","user_id":"3","liked":0,"solved":0,"sender":null,"receiver":null,"intermediate_steps":null,"files":null,"remark":null,"create_time":"2024-08-27T14:01:16","update_time":null,"user_question":null,"knowledge_source":null,"answer_files":null,"id":null,"isSend":true},{"is_bot":true,"source":0,"message":"Which courses did Li Ming study today","extra":"\"{}\"","type":"end_cover","category":"answer","flow_id":"ac5da15cc68f4b289a1123b2486c551f","chat_id":"5f85f74988e830aad4332b09d7f13eab","user_id":"3","liked":0,"solved":0,"sender":null,"receiver":null,"intermediate_steps":null,"files":null,"remark":null,"create_time":"2024-08-27T14:01:16","update_time":null,"user_question":null,"knowledge_source":[],"answer_files":[],"id":"157c3a24049d47f2a51530de80af877e","isSend":false}],"inputs":{"input":"Hello","file_list":[]}}. , Embodiment three:

[0026] In the step 1, the graph link freely defines loop edges and loop nodes in the graph structure, and finally forms a directed acyclic graph. Through the intention recognition capability of the AI ​​basic big model, the connection line is automatically selected, and the agent intelligent body education module components on the execution line are selected. According to different user scenarios, the required big model and different agent intelligent body education module components are selected, and the relationship and task flow between the agent intelligent body modules are defined. The user starts the execution of the graph link through a Request request.

[0027] The sub-agent intelligent body education module includes but is not limited to a teaching auxiliary agent, a question-answering agent, and an evaluation agent. The sub-agent intelligent body education module is classified according to definitions, and different permission access mechanisms are set when setting the classifications. Different sub-agent intelligent body education modules are customized and added according to different permission access mechanisms as needed.

[0028] The repeated ReAct in step 4 includes the following steps: Initialization: The sub-agent education module starts with a certain initial strategy or a completely random behavior pattern; Action: The sub-agent education module selects an action based on the current policy and applies it to the environment; Observation: The environment responds to the actions of the sub-agent education module and produces new states or results. The new state or result information is collected by the sub-agent education module as observations. Evaluation and learning: Based on the results collected from observations, the sub-agent education module evaluates the effectiveness of its actions and updates its own strategy based on the learning algorithm; this step may involve receiving and processing reward signals in order to optimize future decisions; Iteration: The above steps are performed repeatedly. After each cycle, the sub-agent education module may improve its behavior strategy until the preset learning goal or performance standard is reached.

[0029] Among them, the sub-agent intelligent body education module can adjust its own behavior strategy according to environmental feedback, which means that it can maintain effectiveness and adaptability in a constantly changing environment; Through the iterative process, the sub-agent education module can gradually optimize its decision-making ability, improve the efficiency and effectiveness of problem solving, and ultimately achieve the preset learning goals or performance standards; Moreover, each observation provides new information to the sub-agent education module, which is used to evaluate the effectiveness of the current strategy and guide future actions, which ensures that the learning process is based on actual experience rather than purely theoretical assumptions. Reinforcement learning algorithms allow agents to automatically discover which behaviors bring greater rewards and thus optimize themselves, without the need for human intervention to set each specific action. After enough rounds of learning, the sub-agent education module may not only perform well in the training environment, but also apply the knowledge learned in a specific situation to a similar but different new situation, that is, to achieve a certain degree of knowledge transfer. The introduction of reward signals gives the learning process a clear sense of direction, and the sub-agent education module knows which direction to work towards to achieve better results.

[0030] Finally, the above framework can be easily extended to systems containing multiple sub-agent education modules, which can collaborate or compete with each other to complete more complex tasks.

[0031] The learning algorithm in the iterative ReAct adopts the policy gradient method, and the calculation formula of the policy gradient method is: ; in, ∇θ is the strategy parameter; J( θ ) is the objective function, usually expressed as the expected return under the strategy; Eτ~πθ is the expectation operator, the expected value of all possible trajectories τ, where τ is sampled according to the strategy πθ; τ is a complete trajectory, that is τ =(s0,a0,r1,s1,a1,r2,...,sT); πθ(at|st) is the probability of selecting action at according to the current strategy πθ in state st; logπθ(at|st) is the log value of the probability of the strategy choosing an action; ∇θlogπθ(at|st) is the gradient of the log value of the probability of the policy selecting an action with respect to the parameter θ; Rt is The cumulative return starting from time step t; in, ; in, R is the cumulative reward starting from time step t, which represents the sum of all future rewards from the current time step t to the terminal time step T; It represents the accumulation from k=0 to k=Tt, where T is the total number of time steps of the trajectory and t is the current time step. Therefore, Tt represents how many time steps are left from the current time step to the end of the trajectory. γ is the discount factor, and its value range is 0<γ≤1; γk is the discount for future rewards; k is the index of the time step, starting from 0; is the immediate reward obtained at time step t+k+1.

[0032] The above calculation formula directly optimizes the strategy function instead of indirectly influencing the strategy through the value function. The core idea of ​​this type of method is to adjust the strategy parameters so that the actions taken can maximize the cumulative reward.

[0033] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for implementing blueprint orchestration of multiple agents in the AIGC platform education scenario with zero code, characterized by: The method for implementing blueprint orchestration of multiple agents with zero code in the education scenario of the AIGC platform includes the following steps: Step 1: Design a graphical blueprint editing interface to realize the connection and arrangement between the AI ​​basic big model and various agent intelligent body education module components. Users can drag and connect in the blueprint editing interface to form a graph link, output results through the AI ​​big model or execute agent intelligent body education module components; Step 2: predefine the agent module and divide the agent module into multiple sub-agent intelligent education modules according to the multi-faceted definition of the education scenario; Step 3: Define the attributes and behavior rules of the sub-agent intelligent body education module, customize the attributes and behavior rules of the sub-agent intelligent body education module according to the personalized configuration, and the user can realize personalized agent configuration through parameter setting and condition selection; Step 4: Automatically generate execution logic and run to achieve results. Automatically generate the underlying execution logic. According to the line planning connected to the graphical blueprint editing interface and the sub-agent intelligent body education module on the line, rely on the blueprint editor to generate a directed graph, identify the nodes and edge nodes on the directed acyclic graph line in the directed graph, and form text logic with the graph link. The generated text logic is fed to the AI ​​big model, and the AI ​​big model's own intelligent body is used to perform actions and observe environmental reactions, and adjust its subsequent behavior based on these observations, and repeat the ability recognition, interact through the ReAct library, and finally the AI ​​big model returns the result; Step 5: Generate a general protocol interface that can access the graph API based on the orchestrated agent capabilities.

2. According to claim 1, a method for implementing blueprint arrangement of multiple agents in an AIGC platform education scenario with zero code, characterized in that: In the step 1, the graph link freely defines loop edges and loop nodes in the graph structure, and finally forms a directed acyclic graph. Through the intention recognition capability of the AI ​​basic big model, the connection line is automatically selected, and the agent intelligent body education module components on the execution line are selected. According to different user scenarios, the required big model and different agent intelligent body education module components are selected, and the relationship and task flow between the agent intelligent body modules are defined. The user starts the execution of the graph link through a Request request.

3. According to claim 2, a method for implementing blueprint arrangement of multiple agents in an AIGC platform education scenario with zero code, characterized in that: The educational scenario in step 2 includes teacher teaching, student learning, teacher and student evaluation, school administration, and school safety, and defines multiple aspects to divide the agent module into multiple sub-agent intelligent education modules.

4. According to claim 3, a method for implementing blueprint arrangement of multiple agents in an AIGC platform education scenario with zero code, characterized in that: The sub-agent intelligent body education module includes but is not limited to a teaching auxiliary agent, a question-answering agent, and an evaluation agent. The sub-agent intelligent body education module is classified according to definitions, and different permission access mechanisms are set when setting the classifications. Different sub-agent intelligent body education modules are customized and added according to different permission access mechanisms as needed.

5. According to claim 4, a method for implementing blueprint arrangement of multiple agents in an AIGC platform education scenario with zero code, characterized in that: The sub-agent intelligent body education module in the step three is used for agent execution operations, and the agent execution operations are composed of API communication protocols or executable codes. The user sets different parameters and different condition weights as the basis for judgment through the AI ​​basic model.

6. According to claim 5, a method for implementing blueprint arrangement of multiple agents in an AIGC platform education scenario with zero code, characterized in that: In the process of generating the API communication protocol in step five, a unique UUID random string is formed by combining the generated orchestrated graph link, and the effect is achieved through execution logic and operation to realize human-computer interaction. Through the unified API http request protocol, different orchestrated graph links are identified, and then this interaction is initiated. Respond returns the http result, and the selected graph connection is defined in the http parameter to form a proprietary graph API communication protocol interface.

7. According to claim 6, a method for implementing blueprint arrangement of multiple agents in an AIGC platform education scenario with zero code, characterized in that: The repeated ReAct in step 4 includes the following steps: Initialization: The sub-agent education module starts with a certain initial strategy or a completely random behavior pattern; Action: The sub-agent education module selects an action based on the current policy and applies it to the environment; Observation: The environment responds to the actions of the sub-agent education module and produces new states or results. The new state or result information is collected by the sub-agent education module as observations. Evaluation and learning: Based on the results collected from observations, the sub-agent education module evaluates the effectiveness of its actions and updates its own strategy based on the learning algorithm; Iteration: The above steps are performed repeatedly. After each cycle, the sub-agent education module may improve its behavior strategy until the preset learning goal or performance standard is reached.

8. According to claim 7, a method for implementing blueprint arrangement of multiple agents in an AIGC platform education scenario with zero code, characterized in that: The learning algorithm in the iterative ReAct adopts the policy gradient method, and the calculation formula of the policy gradient method is: ； in, ∇θ is the strategy parameter; J( θ ) is the objective function, usually expressed as the expected return under the strategy; Eτ~πθ is the expectation operator, the expected value of all possible trajectories τ, where τ is sampled according to the strategy πθ; τ is a complete trajectory, that is τ =(s0,a0,r1,s1,a1,r2,...,sT); πθ(at|st) is the probability of selecting action at according to the current strategy πθ in state st; logπθ(at|st) is the log value of the probability of the strategy choosing an action; ∇θlogπθ(at|st) is the gradient of the log value of the probability of the policy selecting an action with respect to the parameter θ; Rt is The cumulative return starting from time step t; in, ； in, R is the cumulative reward starting from time step t, which represents the sum of all future rewards from the current time step t to the terminal time step T; It represents the accumulation from k=0 to k=Tt, where T is the total number of time steps of the trajectory and t is the current time step. Therefore, Tt represents how many time steps are left from the current time step to the end of the trajectory. γ is the discount factor, and its value range is 0<γ≤1; γk is the discount for future rewards; k is the index of the time step, starting from 0; is the immediate reward obtained at time step t+k+1.

Images