AI collaborative code generation method based on flow arrangement

Abstract

The invention relates to the technical field of computers, provides an AI collaborative code generation method based on flow arrangement, and innovatively realizes full-flow automation from demand acquisition to complete code engineering through a visual flow arrangement of a low-code platform and a multi-role collaborative mechanism of a large language model. And demand iteration is supported through a dynamic optimization technology. The core objective of the invention is to significantly improve the development efficiency, reduce the technical threshold, and ensure the consistency and expandability in a complex development scene.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of low-code development, artificial intelligence and software engineering, and specifically relates to an AI collaborative code generation method based on a low-code process orchestration platform, which is suitable for efficiently and automatically completing requirement analysis, task decomposition, architecture design, code generation and dynamic iteration optimization in software development. BACKGROUND

[0002] A low-code development platform reduces the complexity of software development through visual process orchestration, but its code generation capability still relies on manual configuration, making it difficult to cope with the automated implementation of complex system requirements. Large language models (LLM) have advantages in code generation, but in existing technologies, the integration of AI and low-code platforms only stops at single-function implementation, lacking full-process collaboration from requirements to complete engineering. In addition, existing methods often need to reconfigure the process when requirements change, which is inefficient and difficult to ensure consistency. Dify and other low-code platforms provide workflow orchestration basics, but do not fully utilize their visual features and AI reasoning capabilities to achieve intelligent code generation and dynamic optimization.

[0003] Therefore, there is an urgent need for an innovative method that combines low-code platforms and AI technology to achieve an efficient, flexible and dynamically optimized code generation process. SUMMARY

[0004] The present application aims to provide an AI collaborative code generation method based on process orchestration, which innovatively realizes full-process automation from requirement collection to complete code engineering through the visual process orchestration of a low-code platform and the multi-role collaboration mechanism of a large language model, and supports requirement iteration through dynamic optimization technology. The core goal of the present application is to significantly improve development efficiency, reduce technical barriers, and ensure consistency and scalability in complex development scenarios.

[0005] To achieve the above technical purposes and effects, the present application discloses an AI collaborative code generation method based on process orchestration, comprising the following steps: One: protocol data flow processing Collect and analyze multi-modal development requirements through a visual dialog box, deploy a visual dialog box component in a low-code process orchestration platform, support users to submit development requirements through multi-modal input such as text, image and voice, and use natural language processing (NLP) and image recognition technology to analyze the semantics of the requirements in combination with a large language model; II. Task decomposition and visual process orchestration Configure AI task decomposition nodes, generate and optimize task decomposition tables, configure AI task decomposition nodes in a low-code platform, based on the structured requirements in step 1, combine the inference capabilities of large language models to automatically generate task decomposition tables, including task name, priority, dependency relationship and estimated working hours, task decomposition results are stored in JSON format and displayed through the platform's drag-and-drop process editor, users can adjust task order or dependency logic in real time; III. Interactive design of product manager node Set up product manager nodes, generate interactive product design documents, configure product manager role nodes, set design instructions through the prompt word editor of the low-code platform, based on the task decomposition table in step two, AI generates product design documents, including functional flowchart, interface prototype and user stories, the platform supports real-time preview and mouse adjustment functions; IV. Intelligent system planning of architect node Set up architect nodes, generate intelligent system architecture design, input technology stack and performance requirements through the parameter configuration interface of the low-code platform, based on the task decomposition in step two and the design document in step three, AI generates system architecture design, including module division, API interface specification, database ER diagram, output in UML diagram and JSON configuration file; V. Automatic code generation of R&D engineer node Set up R&D engineer nodes, generate front-end and back-end code, based on the architecture design in step four, use large language models to generate front-end and back-end code. The front-end code is output in Vue framework, and the back-end code is output in Java format, and the code is detected by the code verification tool of the low-code platform for syntax and specification; VI. Resource integration and engineering automatic generation Integrate the code into a runnable project through Python nodes, configure Python execution nodes, receive the code output in step five, automatically write front-end and back-end code into corresponding files through scripts, and generate standard project engineering, the platform provides one-key export and dependency check functions to ensure that the project can be directly compiled and run; VII. Dynamic optimization and iterative update mechanism Realize demand iteration optimization through dynamic feedback loop, design a dynamic feedback loop in the low-code platform, when users submit demand adjustment or add new functions, the system traces back to step one through the visual interface, combines with the context difference analysis technology, incrementally updates the task decomposition table, design document and code structure, avoids repeated calculation, and finally realizes efficient iteration. Advantages

[0006] 1. Efficiency: Full-process automation reduces development cycle from 30 days in traditional methods to 5 days (based on medium-sized project testing).

[0007] 2. Low threshold: The visual operation of the low-code platform increases the participation rate of non-technical personnel to 70%.

[0008] 3. Flexibility: Supports multiple technology stacks and dynamic demand adjustment, with an application rate covering more than 90% of development scenarios.

[0009] 4. Compared with the mainstream tools in the market: Compared with the mainstream tools in the market (such as GitHub Copilot), the code generated by Copilot needs to be manually configured and compiled, and the engineering conversion time accounts for more than 35% of the total development time (based on industry average data). However, through the automatic resource integration and dependency completion technology of the low-code platform, the engineering time is shortened to less than 5%, and the overall efficiency is significantly improved.

[0010] 5. Innovation: Integration of multi-modal analysis, dynamic task decomposition, self-learning code generation, and differential iteration technology, significantly surpassing existing technologies. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 is the overall flowchart of the present application.

[0012] Figure 2 is the task decomposition and optimization flowchart of the low-code platform in the present application. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application will be further described in detail below in combination with embodiments.

[0014] As shown in Figure 1 , an AI collaborative code generation method based on process arrangement includes the following steps: One: protocol data flow processing Collect and analyze multi-modal development requirements through a visual dialog box, deploy a visual dialog box component in a low-code process arrangement platform, support users to submit development requirements through text, image, and voice multi-modal input, combine a large language model, use natural language processing (NLP) and image recognition technology to analyze the semantics of the requirements; By introducing a deep analysis technology of multi-modal requirements (such as OCR to extract text requirements from images), and through a context semantic association algorithm to improve the consistency of the requirements.

[0015] The technical advantage is that compared with traditional single text input, this method supports diversified requirement collection, and the analysis accuracy is improved to more than 95% (based on test data).

[0016] Two: Task decomposition and visual process orchestration Configure AI task decomposition nodes, generate and optimize task decomposition tables, configure AI task decomposition nodes in low-code platforms, based on the structured requirements in step 1, combine the reasoning capabilities of large language models, automatically generate task decomposition tables, including task name, priority, dependency relationship and estimated working hours, task decomposition results are stored in JSON format, and displayed through the platform's drag-and-drop process editor, users can adjust task order or dependency logic in real time; The innovation lies in proposing a dynamic task dependency analysis algorithm based on graph theory, which automatically optimizes task parallelism and execution order.

[0017] The technical advantage is: task decomposition efficiency is improved by 30%, dependency conflict is reduced to less than 5% (based on a simulation test of 500 tasks).

[0018] Three: Interactive design of product manager node Set the product manager node, generate interactive product design documents, configure the product manager role node, set the design instructions through the prompt word editor of the low-code platform, based on the task decomposition table in step two, AI generates product design documents, including function flowchart, interface prototype and user story, the platform supports real-time preview and mouse adjustment functions; By introducing an adaptive prompt word generator, the prompt content is dynamically optimized according to the task complexity, ensuring the relevance of the design output.

[0019] The technical advantage is: the design document generation time is shortened to 1 / 3 of the original method, and the manual adjustment rate is reduced by 20%.

[0020] Figure 2 The low-code platform task decomposition and optimization flowchart is used in multiple steps.

[0021] Four: Intelligent system planning of architect node Set the architect node, generate intelligent system architecture design, input technology stack and performance requirements through the parameter configuration interface of the low-code platform, based on the task decomposition in step two and the design document in step three, AI generates system architecture design, including module division, API interface specification, database ER diagram, output in UML diagram and JSON configuration file; Adopt integrated architecture mode recommendation engine (such as microservices, layered architecture), automatically match the optimal solution according to the demand size.

[0022] The technical advantage is: the consistency of architecture design reaches 98%, and the manual adjustment of requirements is reduced to less than 10%.

[0023] Five: Automatic code generation for R&D engineer nodes Set up R&D engineer nodes to generate front-end and back-end code based on the architecture design in step four. Use large language models to generate front-end and back-end code. The front-end code is output in the Vue framework, and the back-end code is output in Java format. The code is checked by the code verification tool of the low-code platform to detect syntax and specifications. Propose a code template self-learning mechanism to optimize code structure based on historical generation data and improve maintainability.

[0024] The technical advantage is that the code generation accuracy rate reaches 97%, and the generation speed is improved to 15 seconds per 100 lines of code.

[0025] Six: Resource integration and engineering automation generation Integrate the code into a runnable project through Python nodes. Configure Python execution nodes to receive the code output from step five. Automatically write front-end and back-end code into corresponding files through scripts, and generate standard project engineering. The platform provides one-click export and dependency check functions to ensure that the project can be directly compiled and run. Introduce file dependency auto-completion technology to detect and fix missing dependencies (such as front-end NPM packages).

[0026] The technical advantage is that the project generation success rate reaches 99.5%, and the deployment preparation time is shortened by 50%.

[0027] Seven: Dynamic optimization and iterative update mechanism Realize demand iteration optimization through dynamic feedback loop. Design a dynamic feedback loop process in the low-code platform. When users submit demand adjustments or add new functions, the system traces back to step one through a visual interface, combines context difference analysis technology, and incrementally updates task decomposition table, design document and code structure to avoid repeated calculation, finally realizes efficient iteration. Propose a context difference update algorithm based on version tree to support multi-version demand parallel optimization.

[0028] The technical advantage is that the iteration update efficiency is improved by 40%, and the code redundancy is reduced to less than 3%.

[0029] The following is a specific explanation of developing an "online education platform": The demand is: "Need to support course management, user authentication and score statistics, front-end Vue, back-end Java".

[0030] Step 1: Analyze the demand and generate a structured description (time-consuming 20 seconds).

[0031] Step 2: Decompose into "course management", "user authentication" and "score statistics" three tasks (dependency analysis time-consuming 10 seconds).

[0032] Step 3: Generate course management interface prototype (time-consuming 2 minutes).

[0033] Step 4: Design the architecture, including APIs such as ` / getCourseList` (takes 3 minutes).

[0034] Step 5: Generate approximately 500 lines of Vue+Java code (takes 2 minutes).

[0035] Step 6: Integrate into a Maven project (takes 1 minute).

[0036] Step 7: Add "course search function" and incrementally update the code (takes 3 minutes).

[0037] Result: Total time taken is 11 minutes, a runnable project is generated, and manual intervention is only required to confirm the requirements.

[0038] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A process orchestration-based AI collaborative code generation method, characterized in that: Comprising the following steps: One: protocol data flow processing Collect and analyze multi-modal development requirements through a visual dialog box, deploy the visual dialog box component in a low-code process orchestration platform, support users to submit development requirements through text, image, and voice multi-modal input, combine large language models, use natural language processing (NLP) and image recognition technology to analyze the semantics of the requirements; Two: task decomposition and visual process orchestration Configure AI task decomposition nodes, generate and optimize task decomposition tables, configure AI task decomposition nodes in a low-code platform, based on the structured requirements in step 1, combine the reasoning ability of large language models to automatically generate task decomposition tables, including task name, priority, dependency relationship and estimated working hours, task decomposition results are stored in JSON format, and are displayed through the platform's drag-and-drop process editor, users can adjust the task order or dependency logic in real time; Three: interactive design of product manager node Set up product manager nodes, generate interactive product design documents, configure product manager role nodes, set design instructions through the low-code platform's prompt word editor, based on the task decomposition table in step two, AI generates product design documents, including functional flowchart, interface prototype and user stories, the platform supports real-time preview and mouse adjustment functions; Four: intelligent system planning of architect node Set up architect nodes, generate intelligent system architecture design, input technology stack and performance requirements through the low-code platform's parameter configuration interface, based on the task decomposition in step two and the design document in step three, AI generates system architecture design, including module division, API interface specification, database ER diagram, output in UML diagram and JSON configuration file; Five: automatic code generation of R&D engineer node Set up R&D engineer nodes, generate front-end and back-end code, based on the architecture design in step four, use large language models to generate front-end and back-end code. The front-end code is output in Vue framework, and the back-end code is output in Java format, and the code is detected for syntax and specification through the low-code platform's code verification tool; Six: resource integration and engineering automatic generation Integrate the code into a runnable project through Python nodes, configure Python execution nodes, receive the code output in step five, automatically write front-end and back-end code into corresponding files through scripts, and generate standard project engineering, the platform provides one-key export and dependency check functions to ensure that the project can be directly compiled and run; Seven: dynamic optimization and iterative update mechanism Realize requirement iteration optimization through dynamic feedback loop, design a dynamic feedback loop in the low-code platform, when users submit requirement adjustments or add new functions, the system traces back to step one through the visual interface, combines context difference analysis technology, incrementally updates task decomposition table, design document and code structure, avoids repeated calculation, and finally realizes efficient iteration.

Images