Dynamic document generation method and system based on LLM (Language Language Model)

Abstract

The invention relates to document generation, in particular to a dynamic document generation method and system based on a large language model LLM, and the method comprises the steps: receiving application form data, carrying out the parameter mapping and data binding of the application form data, converting the application form data into structured data, and storing the structured data into a relational database; converting the submitted document into vectorized data, and storing the vectorized data into a vector database; retrieving vectorized data highly related to the submitted document from a vector database to realize deep understanding and intelligent matching of the submitted document; retrieving model configuration information from the relational database, and dynamically creating a model instance according to the model configuration information; according to the variables defined in the document template and the model instances associated with the variables, integrating contents generated by the model instances into the document template so as to specify that specific contents are generated in the document generation process; according to the technical scheme provided by the invention, the defect that the accuracy, the correlation and the diversity of the generated document are relatively insufficient in the prior art can be effectively overcome.

Description

Technical Field

[0001] The present invention relates to document generation, and in particular to a dynamic document generation method and system based on a large language model (LLM). Background Art

[0002] In the field of traditional document generation, fixed templates and static rules are usually used to generate documents. These document generation methods have the defects of insufficient flexibility and poor adaptability, especially when faced with the need to generate complex documents containing multiple types of content (such as UML diagrams, Markdown formatted text, ordinary text, etc.), these traditional methods are often unable to cope with it. In addition, traditional document generation methods cannot generate effective and real documents based on real needs and data assistance, resulting in significant deficiencies in the accuracy and relevance of the generated documents. Summary of the invention

[0003] In view of the above-mentioned shortcomings of the prior art, the present invention provides a dynamic document generation method and system based on a large language model (LLM), which can effectively overcome the defects of the prior art in that the accuracy, relevance and diversity of the generated documents are greatly insufficient.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: The dynamic document generation method based on the large language model LLM includes the following steps: S1. User submits documents and application form; S2. Receive application form data, perform parameter mapping and data binding on the application form data, convert the application form data into structured data, and store it in a relational database; S3, convert the submitted document into vectorized data and store it in the vector database; S4. Use the retrieval-augmented generation model RAG to retrieve vectorized data that is highly relevant to the submitted document from the vector database, and achieve in-depth understanding and intelligent matching of the submitted document; S5. Retrieve model configuration information from a relational database, and dynamically create a model instance according to the model configuration information; S6. Integrate the content generated by the model instance into the document template according to the variables defined in the document template and their associated model instances, so as to specify the content generated during the document generation process; S7, dynamically render variables on the document template, and render the content generated by the model instance in the document template into multiple types of content; S8. Make fine adjustments and verifications on the rendered document template to generate the final document.

[0005] Preferably, receiving the application form data in S2, performing parameter mapping and data binding on the application form data, converting the application form data into structured data, and storing the data in a relational database includes: Receive application form data through the RESTful API interface and convert the application form data into structured data through parameter mapping. In this process, seamless mapping between user input and the system's internal data structure is achieved through data binding to ensure data accuracy and consistency; Persistently store structured data in a relational database as a single task record for subsequent automatic retry and data management operations; Among them, the relational database stores large language models LLM, structured data, model configuration information, document templates and variable configuration information, and is maintained through strict and complete lifecycle management to ensure the efficiency and consistency of model dynamic instantiation and configuration information management, as well as the consistency and updatability of document templates, thereby improving the maintainability and stability of the system.

[0006] Preferably, S3 converts the submitted document into vectorized data and stores it in a vector database, including: Perform data cleaning on submitted documents, convert pre-processed document data into vectorized data using vector space model (VSM), and store it in vector database; Use the vector space model VSM to efficiently index vectorized data to achieve fast retrieval and intelligent matching; During the data storage process, a distributed storage architecture is adopted to improve the reliability and scalability of data storage.

[0007] Preferably, S4 uses the retrieval enhancement generation model RAG to retrieve vectorized data highly related to the submitted document from the vector database to achieve in-depth understanding and intelligent matching of the submitted document, including: Utilize the context-awareness of the retrieval-augmented generation model RAG to retrieve vectorized data that is highly relevant to the submitted document from the vector database; Based on the complex neural network architecture of the Retrieval Enhanced Generation Model (RAG), we can achieve deep understanding and intelligent matching of submitted documents, and improve the accuracy and relevance of document generation.

[0008] Preferably, in S5, retrieving the model configuration information from the relational database and dynamically creating the model instance according to the model configuration information includes: The ModelInfoService class retrieves model configuration information from a relational database, dynamically creates a model instance based on the model configuration information, and implements full-process management of the model life cycle to ensure the interchangeability and compatibility of the model. Through dependency injection technology, the creation and configuration of dynamic model instantiation is realized to improve the flexibility and maintainability of the system; Among them, the model instance dynamically created by the ModelInfoService class according to the model configuration information is the large language model LLM. The ModelInfoService class provides a unified AIModel model calling interface, which encapsulates the standard method of model calling and supports complex text and image input.

[0009] Preferably, in S6, according to the variables defined in the document template and the model instances associated therewith, the content generated by the model instances is integrated into the document template to specify the content generated during the document generation process, including: The created model instances are called sequentially according to the variable configuration information to generate content, and the placeholders in the document template are replaced with the content generated by the corresponding model instances. The content generated by the model instances is integrated into the document template to specify the content generated during the document generation process; There is a fault-tolerance mechanism in the process of calling the created model instance to ensure automatic retries when a failure occurs during the model call process until the model call succeeds or the maximum number of attempts is reached.

[0010] Preferably, in S7, dynamic variable rendering is performed on the document template, and the content generated by the model instance in the document template is rendered into multiple types of content, including: Dynamically render variables in document templates, and render the content generated by model instances in document templates into multiple types of content including UML diagrams, Markdown formatted text, and ordinary text, to ensure the diversity and professionalism of document generation; During the process of dynamic variable rendering of the document template, each placeholder in the document template is checked. For the placeholders that have not been processed, the corresponding service is called to obtain the content, and the corresponding rendering processing is performed based on the obtained content. If any exception is encountered, an exception is thrown for error handling to ensure the stability and reliability of document generation.

[0011] Preferably, in S8, fine-tuning and verifying the rendered document template to generate a final document includes: The template rendering engine technology is used to dynamically fill the processed data into the corresponding placeholders according to the predefined document template structure to realize the automatic construction of document content. The template rendering engine technology supports complex document element operations including paragraph typesetting, style application and image embedding, ensuring that the generated document meets professional standards visually. Make fine adjustments to document templates, including font specifications, spacing adjustments, and header and footer settings, to ensure that the generated documents meet industry standards and meet user personalized needs; Verify the structure and content of the document template to ensure the completeness and accuracy of the generated document.

[0012] Preferably, the method further comprises: During the document generation process, the exception monitoring mechanism records the abnormal situation. Any exceptions that occur during data preprocessing, model calling, and dynamic variable rendering will be captured and recorded in the log file. The log analysis technology can be used to quickly locate and respond to the exceptions. Dynamic optimization including adjusting database indexing strategies, optimizing algorithm efficiency and adjusting model parameters is performed according to actual operating conditions to improve the accuracy and efficiency of document generation. At the same time, system performance is continuously improved through performance monitoring and optimization technologies.

[0013] A dynamic document generation system based on a large language model (LLM), including a user interface layer, a data processing layer, a storage layer, a model processing layer, a document generation layer, a monitoring layer, and an output layer; The user interface layer, where users submit documents and application forms; The data processing layer receives the application form data, performs parameter mapping and data binding on the application form data, converts the application form data into structured data, and stores it in a relational database. It also converts the submitted documents into vectorized data and stores it in a vector database. The storage layer includes a relational database for storing structured data and a vector database for storing vectorized data. The model processing layer uses the retrieval enhanced generation model RAG to retrieve vectorized data highly related to the submitted document from the vector database to achieve in-depth understanding and intelligent matching of the submitted document, retrieve the model configuration information from the relational database, dynamically create a model instance based on the model configuration information, and integrate the content generated by the model instance into the document template based on the variables defined in the document template and its associated model instance to specify the content generated during the document generation process; The document generation layer performs dynamic variable rendering on the document template, renders the content generated by the model instance in the document template into multiple types of content, and makes fine adjustments and verifications on the rendered document template to generate the final document. Monitoring layer: During the document generation process, abnormal situations are recorded through the abnormal monitoring mechanism and recorded in the log file. The log analysis technology is used to quickly locate and respond to abnormalities. At the same time, dynamic optimization is performed according to the actual operation of the system, and the system performance is continuously improved through performance monitoring and optimization technology. Output layer, outputs the final document.

[0014] Compared with the prior art, the dynamic document generation method and system based on the large language model LLM provided by the present invention has the following beneficial effects: 1) Use the retrieval-augmented generation model RAG to perform in-depth search and analysis on the vectorized data stored in the vector database to assist in document generation. The retrieval-augmented generation model RAG can deeply understand the contextual information of the submitted document and improve the accuracy and relevance of document generation; 2) A multi-model access and model dynamic instantiation mechanism is proposed, which can dynamically create model instances according to the model configuration information in the relational database, realize the full process management of the model life cycle, and provide a unified AIModel model calling interface, which enhances the flexibility and scalability of the system; 3) Allows definition of variables corresponding to placeholders in document templates. These variables can be associated with specific AI models and set model roles to specify specific content generated during document generation, thus achieving dynamic configuration and intelligent processing of variables; 4) Through the configuration request intelligent processing mechanism, the system can replace the placeholders in the document template with the content generated by the corresponding AI model, realizing the dynamic configuration of the document template and the personalized generation of the document content; 5) Supports rendering the content generated by the model instance in the document template into multiple types of content including UML diagrams, Markdown formatted text and ordinary text, and can flexibly generate professional-level documents according to user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 It is a process and system schematic diagram of the present invention; Figure 2 This is a system interface diagram of the document template in the present invention; Figure 3 This is a system interface diagram for variable configuration information in the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0018] Dynamic document generation methods based on large language models LLM, such as Figure 1 As shown, S1, the user submits documents and application form.

[0019] S2. Receive application form data, perform parameter mapping and data binding on the application form data, convert the application form data into structured data, and store it in a relational database, including: Receive application form data through the RESTful API interface and convert the application form data into structured data through parameter mapping. In this process, seamless mapping between user input and the system's internal data structure is achieved through data binding to ensure data accuracy and consistency; Persistently store structured data in a relational database as a single task record for subsequent automatic retry and data management operations; Among them, the relational database stores large language models LLM, structured data, model configuration information, document templates and variable configuration information, and is maintained through strict and complete lifecycle management to ensure the efficiency and consistency of model dynamic instantiation and configuration information management, as well as the consistency and updatability of document templates, thereby improving the maintainability and stability of the system.

[0020] S3: Convert the submitted document into vectorized data and store it in the vector database, including: Perform data cleaning on submitted documents, convert pre-processed document data into vectorized data using vector space model (VSM), and store it in vector database; Use the vector space model VSM to efficiently index vectorized data to achieve fast retrieval and intelligent matching; During the data storage process, a distributed storage architecture is adopted to improve the reliability and scalability of data storage.

[0021] S4. Retrieve vectorized data highly relevant to the submitted document from the vector database using the retrieval-augmented generation model RAG to achieve in-depth understanding and intelligent matching of the submitted document, including: Utilize the context-awareness of the retrieval-augmented generation model RAG to retrieve vectorized data that is highly relevant to the submitted document from the vector database; Based on the complex neural network architecture of the Retrieval Enhanced Generation Model (RAG), we can achieve deep understanding and intelligent matching of submitted documents, and improve the accuracy and relevance of document generation.

[0022] S5. Retrieve model configuration information from a relational database and dynamically create a model instance based on the model configuration information, specifically including: The ModelInfoService class retrieves model configuration information from a relational database, dynamically creates a model instance based on the model configuration information, implements full-process management of the model life cycle, and ensures the interchangeability and compatibility of the model. Through dependency injection technology, the creation and configuration of dynamic model instantiation is realized to improve the flexibility and maintainability of the system; Among them, the model instance dynamically created by the ModelInfoService class according to the model configuration information is the large language model LLM. The ModelInfoService class provides a unified AIModel model calling interface, which encapsulates the standard method of model calling and supports complex text and image input.

[0023] The system interface diagram of the model configuration information of the present invention can add a large language model LLM that is the same as the content generation, wherein all models can be used for variable association calling, and the relationship between the models and the variables is many-to-many.

[0024] S6. Integrate the content generated by the model instance into the document template according to the variables defined in the document template and their associated model instances, so as to specify the content generated during the document generation process, including: The created model instances are called sequentially according to the variable configuration information to generate content, and the placeholders in the document template are replaced with the content generated by the corresponding model instances. The content generated by the model instances is integrated into the document template to specify the content generated during the document generation process; There is a fault-tolerance mechanism in the process of calling the created model instance to ensure automatic retries when a failure occurs during the model call process until the model call succeeds or the maximum number of attempts is reached.

[0025] Figure 2 This is a system interface diagram of the document template in the present invention. The interface diagram supports the configuration of multiple document templates, which are the basis for generating the final document.

[0026] S7. Dynamically render the document template with variables, and render the content generated by the model instance in the document template into multiple types of content, including: Dynamically render variables in document templates, and render the content generated by model instances in document templates into multiple types of content including UML diagrams, Markdown formatted text, and ordinary text, to ensure the diversity and professionalism of document generation; During the process of dynamic variable rendering of the document template, each placeholder in the document template is checked. For the placeholders that have not been processed, the corresponding service is called to obtain the content, and the corresponding rendering processing is performed based on the obtained content. If any exception is encountered, an exception is thrown for error handling to ensure the stability and reliability of document generation.

[0027] Figure 3 This is a system interface diagram of the variable configuration information in the present invention. This interface diagram supports the selection of the created AI model and the type of final output to the document template: including text, pictures, flowcharts, charts, formulas, etc. When dynamically rendering variables in a document template, the variable is first obtained according to the placeholder, and the prompt words and specific content required for the variable are obtained through the variable scheduling engine. The created AI model is called to generate content according to the variable configuration information, and then the content generated by the AI ​​model is rendered according to the type that is finally output to the document template.

[0028] The specific process of dynamic variable rendering of document templates is as follows: The document template contains placeholders such as {{name}} and {{system description}}. The system automatically identifies and extracts these placeholders from the document template; According to the name of the extracted placeholder, the system automatically matches the corresponding document variables and performs type analysis on these variables through the variable scheduling engine to identify the required prompt words and specific content; The system constructs contextual information related to the variables based on the pre-set prompt word engineering rules and passes this information to the AI ​​model to ensure that the AI ​​model can accurately understand and generate the required content; The AI ​​model generates high-quality text content based on the received contextual information and prompt words, and fills it into the document template after rendering.

[0029] S8. Make fine adjustments and verifications on the rendered document template to generate the final document, including: The template rendering engine technology is used to dynamically fill the processed data into the corresponding placeholders according to the predefined document template structure to realize the automatic construction of document content. The template rendering engine technology supports complex document element operations including paragraph typesetting, style application and image embedding, ensuring that the generated document meets professional standards visually. Make fine adjustments to document templates, including font specifications, spacing adjustments, and header and footer settings, to ensure that the generated documents meet industry standards and meet user personalized needs; Verify the structure and content of the document template to ensure the completeness and accuracy of the generated document.

[0030] In the technical solution of this application, the dynamic document generation method based on the large language model LLM also includes: During the document generation process, the exception monitoring mechanism records the abnormal situation. Any exceptions that occur during data preprocessing, model calling, and dynamic variable rendering will be captured and recorded in the log file. The log analysis technology can be used to quickly locate and respond to the exceptions. Dynamic optimization including adjusting database indexing strategies, optimizing algorithm efficiency and adjusting model parameters is performed according to actual operating conditions to improve the accuracy and efficiency of document generation. At the same time, system performance is continuously improved through performance monitoring and optimization technologies.

[0031] Based on the above-disclosed method for generating dynamic documents based on a large language model LLM, the technical solution of this application also discloses a system for generating dynamic documents based on a large language model LLM, such as Figure 1 As shown, it includes user interface layer, data processing layer, storage layer, model processing layer, document generation layer, monitoring layer and output layer; The user interface layer, where users submit documents and application forms; The data processing layer receives the application form data, performs parameter mapping and data binding on the application form data, converts the application form data into structured data, and stores it in a relational database. It also converts the submitted documents into vectorized data and stores it in a vector database. The storage layer includes a relational database for storing structured data and a vector database for storing vectorized data. The model processing layer uses the retrieval enhanced generation model RAG to retrieve vectorized data highly related to the submitted document from the vector database to achieve in-depth understanding and intelligent matching of the submitted document, retrieve the model configuration information from the relational database, dynamically create a model instance based on the model configuration information, and integrate the content generated by the model instance into the document template based on the variables defined in the document template and its associated model instance to specify the content generated during the document generation process; The document generation layer performs dynamic variable rendering on the document template, renders the content generated by the model instance in the document template into multiple types of content, and makes fine adjustments and verifications on the rendered document template to generate the final document. Monitoring layer: During the document generation process, abnormal situations are recorded through the abnormal monitoring mechanism and recorded in the log file. The log analysis technology is used to quickly locate and respond to abnormalities. At the same time, dynamic optimization is performed according to the actual operation of the system, and the system performance is continuously improved through performance monitoring and optimization technology. Output layer, outputs the final document.

[0032] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic document generation method based on a large language model (LLM), characterized by: The following steps are involved: S1. User submits documents and application form; S2. Receive application form data, perform parameter mapping and data binding on the application form data, convert the application form data into structured data, and store it in a relational database; S3, convert the submitted document into vectorized data and store it in the vector database; S4. Use the retrieval-augmented generation model RAG to retrieve vectorized data that is highly relevant to the submitted document from the vector database, and achieve in-depth understanding and intelligent matching of the submitted document; S5. Retrieve model configuration information from a relational database, and dynamically create a model instance according to the model configuration information; S6. Integrate the content generated by the model instance into the document template according to the variables defined in the document template and their associated model instances, so as to specify the content generated during the document generation process; S7, dynamically render variables on the document template, and render the content generated by the model instance in the document template into multiple types of content; S8. Make fine adjustments and verifications on the rendered document template to generate the final document.

2. The method for generating dynamic documents based on a large language model (LLM) according to claim 1, characterized in that: S2 receives the application form data, performs parameter mapping and data binding on the application form data, converts the application form data into structured data, and stores it in a relational database, including: Receive application form data through the RESTful API interface and convert the application form data into structured data through parameter mapping. In this process, seamless mapping between user input and the system's internal data structure is achieved through data binding to ensure data accuracy and consistency; Persistently store structured data in a relational database as a single task record for subsequent automatic retry and data management operations; Among them, the relational database stores large language models LLM, structured data, model configuration information, document templates and variable configuration information, and is maintained through strict and complete lifecycle management to ensure the efficiency and consistency of model dynamic instantiation and configuration information management, as well as the consistency and updatability of document templates, thereby improving the maintainability and stability of the system.

3. The method for generating dynamic documents based on a large language model (LLM) according to claim 2, characterized in that: S3 converts submitted documents into vectorized data and stores them in the vector database, including: Perform data cleaning on submitted documents, convert pre-processed document data into vectorized data using vector space model (VSM), and store it in vector database; Use the vector space model VSM to efficiently index vectorized data to achieve fast retrieval and intelligent matching; During the data storage process, a distributed storage architecture is adopted to improve the reliability and scalability of data storage.

4. The method for generating dynamic documents based on a large language model (LLM) according to claim 3, characterized in that: S4 uses the retrieval-augmented generation model RAG to retrieve vectorized data that is highly relevant to the submitted document from the vector database, achieving in-depth understanding and intelligent matching of the submitted document, including: Utilize the context-awareness of the retrieval-augmented generation model RAG to retrieve vectorized data that is highly relevant to the submitted document from the vector database; Based on the complex neural network architecture of the Retrieval Enhanced Generation Model (RAG), we can achieve deep understanding and intelligent matching of submitted documents, and improve the accuracy and relevance of document generation.

5. The method for generating dynamic documents based on a large language model (LLM) according to claim 4, characterized in that: S5 retrieves model configuration information from the relational database and dynamically creates a model instance based on the model configuration information, including: The ModelInfoService class retrieves model configuration information from a relational database, dynamically creates a model instance based on the model configuration information, and implements full-process management of the model life cycle to ensure the interchangeability and compatibility of the model. Through dependency injection technology, the creation and configuration of dynamic model instantiation is realized to improve the flexibility and maintainability of the system; Among them, the model instance dynamically created by the ModelInfoService class according to the model configuration information is the large language model LLM. The ModelInfoService class provides a unified AIModel model calling interface, which encapsulates the standard method of model calling and supports complex text and image input.

6. The method for generating dynamic documents based on a large language model (LLM) according to claim 5, characterized in that: In S6, based on the variables defined in the document template and their associated model instances, the content generated by the model instances is integrated into the document template to specify the content generated during the document generation process, including: The created model instances are called sequentially according to the variable configuration information to generate content, and the placeholders in the document template are replaced with the content generated by the corresponding model instances. The content generated by the model instances is integrated into the document template to specify the content generated during the document generation process; There is a fault-tolerance mechanism in the process of calling the created model instance to ensure automatic retries when a failure occurs during the model call process until the model call succeeds or the maximum number of attempts is reached.

7. The method for generating dynamic documents based on a large language model (LLM) according to claim 6, characterized in that: In S7, dynamic variable rendering is performed on the document template, and the content generated by the model instance in the document template is rendered into multiple types of content, including: Dynamically render variables in document templates, and render the content generated by model instances in document templates into multiple types of content including UML diagrams, Markdown formatted text, and ordinary text, to ensure the diversity and professionalism of document generation; During the process of dynamic variable rendering of the document template, each placeholder in the document template is checked. For the placeholders that have not been processed, the corresponding service is called to obtain the content, and the corresponding rendering processing is performed based on the obtained content. If any exception is encountered, an exception is thrown for error handling to ensure the stability and reliability of document generation.

8. The method for generating dynamic documents based on a large language model (LLM) according to claim 7, characterized in that: In S8, the rendered document template is fine-tuned and verified to generate the final document, including: The template rendering engine technology is used to dynamically fill the processed data into the corresponding placeholders according to the predefined document template structure to realize the automatic construction of document content. The template rendering engine technology supports complex document element operations including paragraph typesetting, style application and image embedding, ensuring that the generated document meets professional standards visually. Make fine adjustments to document templates, including font specifications, spacing adjustments, and header and footer settings, to ensure that the generated documents meet industry standards and meet user personalized needs; Verify the structure and content of the document template to ensure the completeness and accuracy of the generated document.

9. The method for generating dynamic documents based on a large language model (LLM) according to any one of claims 1 to 8, characterized in that: The method further includes: During the document generation process, the exception monitoring mechanism records the abnormal situation. Any exceptions that occur during data preprocessing, model calling, and dynamic variable rendering will be captured and recorded in the log file. The log analysis technology can be used to quickly locate and respond to the exceptions. Dynamic optimization including adjusting database indexing strategies, optimizing algorithm efficiency and adjusting model parameters is performed according to actual operating conditions to improve the accuracy and efficiency of document generation. At the same time, system performance is continuously improved through performance monitoring and optimization technologies.

10. A dynamic document generation system based on a large language model (LLM), used to execute the dynamic document generation method based on a large language model (LLM) according to claim 9, characterized in that: It includes user interface layer, data processing layer, storage layer, model processing layer, document generation layer, monitoring layer and output layer; The user interface layer, where users submit documents and application forms; The data processing layer receives the application form data, performs parameter mapping and data binding on the application form data, converts the application form data into structured data, and stores it in a relational database. It also converts the submitted documents into vectorized data and stores it in a vector database. The storage layer includes a relational database for storing structured data and a vector database for storing vectorized data. The model processing layer uses the retrieval enhanced generation model RAG to retrieve vectorized data highly related to the submitted document from the vector database to achieve in-depth understanding and intelligent matching of the submitted document, retrieve the model configuration information from the relational database, dynamically create a model instance based on the model configuration information, and integrate the content generated by the model instance into the document template based on the variables defined in the document template and its associated model instance to specify the content generated during the document generation process; The document generation layer performs dynamic variable rendering on the document template, renders the content generated by the model instance in the document template into multiple types of content, and makes fine adjustments and verifications on the rendered document template to generate the final document. Monitoring layer: During the document generation process, abnormal situations are recorded through the abnormal monitoring mechanism and recorded in the log file. The log analysis technology is used to quickly locate and respond to abnormalities. At the same time, dynamic optimization is performed according to the actual operation of the system, and the system performance is continuously improved through performance monitoring and optimization technology. Output layer, outputs the final document.

Images