Large model-based code intelligence analysis and auditing system and method

Abstract

The application discloses a code intelligent analysis and auditing system and method based on a large language model, and belongs to the technical field of software engineering and artificial intelligence. The system comprises a front-end display layer, a core business processing layer and a data persistence layer. The core business processing layer is provided with a multi-source project obtainer, an intelligent code auditing engine, a multi-model collaborative scheduling and fusion engine and a task scheduler. The method imports code through the multi-source project obtainer, utilizes the multi-model collaborative scheduling and fusion engine to schedule multiple large language models in parallel to analyze code units and fuse results, and stores the results after multi-dimensional auditing by the intelligent code auditing engine. The application solves the problems of insufficient semantic understanding of traditional tools and low accuracy of general large models in specific project analysis, effectively improves the accuracy, efficiency and automation level of code auditing, and supports the generation of professional auditing reports.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of software engineering and artificial intelligence, and more specifically, to a system and method for intelligent code analysis and auditing based on large models. Background Technology

[0002] Traditional programming aids and static code analysis tools primarily rely on predefined syntax rules and pattern matching to provide code completion or basic error detection. However, these tools have significant shortcomings in deep semantic understanding, context awareness, and multi-objective optimization combined with specific business logic. They struggle to understand the mapping relationship between the code's true intent and business requirements, and cannot effectively identify complex logical defects and functional vulnerabilities that require consideration of project context and business scenarios, resulting in limited analysis results.

[0003] In recent years, with the development of large language model technology, it has shown great potential in code generation and understanding, providing a new path for intelligent code analysis. However, directly applying general-purpose large language models to code auditing of specific projects still faces many challenges: the models lack a deep understanding of the high-level architecture and business domain knowledge of specific projects, resulting in low accuracy and high false positive and false negative rates in complex scenarios; at the same time, existing solutions have not yet formed systematic solutions in terms of analysis mode flexibility, multi-model collaboration, contextual information integration, and continuous evolution of audit knowledge, making it difficult to meet the high standards of accuracy, efficiency, and interpretability required by enterprise-level code auditing. Summary of the Invention

[0004] To address the aforementioned technical problems in related technologies, this invention proposes a code intelligent analysis and auditing system and method based on a large model, which can overcome the above-mentioned shortcomings of existing technologies.

[0005] To achieve the above-mentioned technical objectives, the technical solution of the present invention is implemented as follows: A code intelligent analysis and auditing system based on large models; This large-model-based intelligent code analysis and auditing system includes: The front-end presentation layer is used to provide the user interface and includes at least a project management module, an audit task module, and a code analysis module. The core business processing layer, which communicates with the front-end presentation layer, includes: A multi-source project acquirer to support importing code projects from remote repositories or local compressed files; The intelligent code auditing engine is used to perform multi-dimensional analysis on imported code projects. The multi-dimensional analysis includes at least security vulnerability detection, performance analysis, and maintainability assessment. A multi-model collaborative scheduling and fusion engine is used to schedule multiple heterogeneous large language models to perform parallel analysis on the same code unit, and to fuse and arbitrate the output results of each model to generate a unified analysis conclusion. The task scheduler is used to manage the execution of audit tasks; The data persistence layer is used to store system data, including at least an audit issue table for storing issues discovered during code analysis.

[0006] Furthermore, the multi-source project acquirer includes: GitHub API processors and / or GitLab API processors are used to retrieve project data from the corresponding remote repositories; A ZIP file processor used to decompress local compressed files and extract project data.

[0007] Furthermore, the multi-model collaborative scheduling and fusion engine includes: The parallel scheduling module is used to concurrently submit analysis requests for the same code unit to multiple preset large language model services; The results aggregation and arbitration module is used to perform structured parsing of the raw analysis results returned by each model, and to perform weighted voting and semantic fusion based on preset weights and confidence levels. For conflicting analysis results, meta-model arbitration or rule base verification is triggered to generate a final conclusion.

[0008] Furthermore, the core business processing layer also includes context-aware prompting engineering services and a dynamic software bill of materials analysis engine; The dynamic software bill of materials analysis engine is used to parse project dependency files, generate a software bill of materials in a standard format, and match it with external vulnerability intelligence sources to construct a project security dependency graph. The context-aware prompting engineering service is used to integrate code context, the security dependency graph, and project historical issue information before the multi-model collaborative scheduling and fusion engine scheduling of the large language model, to construct enhanced prompt words, and dynamically inject the associated vulnerability information in the security dependency graph into the enhanced prompt words.

[0009] Furthermore, the data persistence layer also includes an audit knowledge graph library, which includes: Issue pattern table, used to store the characteristic fingerprints of code issues and fix templates; The case sample table is used to store positive and negative samples that have been manually verified. The feedback and learning record form is used to record human feedback on the analysis results and the adjustments made by the system accordingly. The project audit history snapshot table is used to store the code status hash and audit result snapshots at the time of each audit to support incremental analysis.

[0010] Furthermore, the task scheduler is also used to drive an incremental-knowledge dual-loop audit process, including: The incremental analysis loop is used to trigger an incremental scan by comparing the project audit history snapshot table after code changes. The knowledge evolution loop is used to structurally store each audit result in the audit knowledge graph library, and dynamically optimize the prompt words submitted to the large language model based on the accumulated case samples and problem patterns.

[0011] Furthermore, the code analysis module in the front-end presentation layer includes: The real-time code analysis component is used to receive code snippets input by the user and trigger real-time analysis; The file selector component is used to browse and select files to be analyzed within the project file; An analysis results visualization component used to display code analysis results.

[0012] Furthermore, the system also includes: The API gateway layer is set between the front-end presentation layer and the core business processing layer to provide standardized application programming interface routing and request processing. The external service integration layer is used to integrate and provide a unified interface for interacting with GitHub and GitLab services.

[0013] According to another aspect of the present invention, a method for intelligent code analysis and auditing based on a large model is provided; This code intelligent analysis and auditing method based on a large model includes the following steps: Import code projects using the multi-source project acquirer; Create an audit task for the code project using the task scheduler; The intelligent code auditing engine performs multi-dimensional analysis of the code files in the code project; During the analysis process, multiple heterogeneous large language models are scheduled in parallel to analyze the same code unit through multi-model collaborative scheduling and fusion engine scheduling, and the analysis results are fused and arbitrated to generate a unified conclusion. Issues identified during the analysis are stored in the audit issue table of the data persistence layer.

[0014] Furthermore, the steps of using multi-model collaborative scheduling and fusion engine scheduling to perform parallel analysis of the same code unit using multiple heterogeneous large language models, and then fusing and arbitrating the analysis results to generate a unified conclusion include: Concurrently submit analysis requests to multiple large language model services; Receive the analysis results returned by each model and perform structured parsing; The parsed results are weighted and fused semantically based on preset model weights, historical accuracy, and real-time confidence. If there is a conflict in the voting results, the meta-model arbitration or rule base verification process will be triggered to generate the final analysis conclusion.

[0015] The beneficial effects of this invention are as follows: By introducing a multi-model collaborative scheduling and fusion mechanism, combined with context-aware prompting engineering and dynamic security dependency graphs, the system's understanding and analysis accuracy of deep code semantics and project-specific contexts are significantly improved, thereby effectively increasing the detection rate and accuracy of code security vulnerabilities and quality defects, while reducing false alarms; furthermore, by constructing an evolvable audit knowledge base and intelligent task scheduling, continuous optimization of analytical capabilities and a significant improvement in audit efficiency are achieved, realizing the integrated intelligent audit goal of supporting multi-source projects and providing professional reports. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is an architecture diagram of a code intelligent analysis and auditing system based on a large model, as described in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0019] like Figure 1 As shown, the code intelligent analysis and auditing system based on a large model according to an embodiment of the present invention includes: The front-end presentation layer is used to provide the user interface and includes at least a project management module, an audit task module, and a code analysis module. The core business processing layer, which communicates with the front-end presentation layer, includes: A multi-source project acquirer to support importing code projects from remote repositories or local compressed files; The intelligent code auditing engine is used to perform multi-dimensional analysis on imported code projects. The multi-dimensional analysis includes at least security vulnerability detection, performance analysis, and maintainability assessment. A multi-model collaborative scheduling and fusion engine is used to schedule multiple heterogeneous large language models to perform parallel analysis on the same code unit, and to fuse and arbitrate the output results of each model to generate a unified analysis conclusion. The task scheduler is used to manage the execution of audit tasks; The data persistence layer is used to store system data, including at least an audit issue table for storing issues discovered during code analysis.

[0020] According to an embodiment of the code intelligent analysis and auditing system based on a large model described in this invention, in a specific implementation, the multi-source project acquirer includes: GitHub API processors and / or GitLab API processors are used to retrieve project data from the corresponding remote repositories; A ZIP file processor used to decompress local compressed files and extract project data.

[0021] According to an embodiment of the code intelligent analysis and auditing system based on a large model as described in this invention, in a specific implementation, the multi-model collaborative scheduling and fusion engine includes: The parallel scheduling module is used to concurrently submit analysis requests for the same code unit to multiple preset large language model services; The results aggregation and arbitration module is used to perform structured parsing of the raw analysis results returned by each model, and to perform weighted voting and semantic fusion based on preset weights and confidence levels. For conflicting analysis results, meta-model arbitration or rule base verification is triggered to generate a final conclusion.

[0022] In a specific embodiment of the code intelligent analysis and auditing system based on a large model according to an embodiment of the present invention, the core business processing layer further includes a context-aware prompting engineering service and a dynamic software bill of materials analysis engine. The dynamic software bill of materials analysis engine is used to parse project dependency files, generate a software bill of materials in a standard format, and match it with external vulnerability intelligence sources to construct a project security dependency graph. The context-aware prompting engineering service is used to integrate code context, the security dependency graph, and project historical issue information before the multi-model collaborative scheduling and fusion engine scheduling of the large language model, to construct enhanced prompt words, and dynamically inject the associated vulnerability information in the security dependency graph into the enhanced prompt words.

[0023] According to an embodiment of the code intelligent analysis and auditing system based on a large model described in this invention, in a specific implementation, the data persistence layer further includes an audit knowledge graph library, which includes: Issue pattern table, used to store the characteristic fingerprints of code issues and fix templates; The case sample table is used to store positive and negative samples that have been manually verified. The feedback and learning record form is used to record human feedback on the analysis results and the adjustments made by the system accordingly. The project audit history snapshot table is used to store the code status hash and audit result snapshots at the time of each audit to support incremental analysis.

[0024] In a specific embodiment of the code intelligent analysis and auditing system based on a large model according to an embodiment of the present invention, the task scheduler is further used to drive an incremental-knowledge dual-loop auditing process, including: The incremental analysis loop is used to trigger an incremental scan by comparing the project audit history snapshot table after code changes. The knowledge evolution loop is used to structurally store each audit result in the audit knowledge graph library, and dynamically optimize the prompt words submitted to the large language model based on the accumulated case samples and problem patterns.

[0025] According to an embodiment of the code intelligent analysis and auditing system based on a large model described in this invention, in a specific implementation, the code analysis module in the front-end presentation layer includes: The real-time code analysis component is used to receive code snippets input by the user and trigger real-time analysis; The file selector component is used to browse and select files to be analyzed within the project file; An analysis results visualization component used to display code analysis results.

[0026] In a specific embodiment of the code intelligent analysis and auditing system based on a large model according to an embodiment of the present invention, the system further includes: The API gateway layer is set between the front-end presentation layer and the core business processing layer to provide standardized application programming interface routing and request processing. The external service integration layer is used to integrate and provide a unified interface for interacting with GitHub and GitLab services.

[0027] Secondly, according to the embodiment of the present invention, the code intelligent analysis and auditing method based on a large model includes the following steps: Import code projects using the multi-source project acquirer; Create an audit task for the code project using the task scheduler; The intelligent code auditing engine performs multi-dimensional analysis of the code files in the code project; During the analysis process, multiple heterogeneous large language models are scheduled in parallel to analyze the same code unit through multi-model collaborative scheduling and fusion engine scheduling, and the analysis results are fused and arbitrated to generate a unified conclusion. Issues identified during the analysis are stored in the audit issue table of the data persistence layer.

[0028] According to an embodiment of the present invention, the method for intelligent code analysis and auditing based on large models, in a specific implementation, includes the following steps: scheduling multiple heterogeneous large language models in parallel to analyze the same code unit through multi-model collaborative scheduling and fusion engine scheduling, and then fusion and arbitrating the analysis results to generate a unified conclusion. Concurrently submit analysis requests to multiple large language model services; Receive the analysis results returned by each model and perform structured parsing; The parsed results are weighted and fused semantically based on preset model weights, historical accuracy, and real-time confidence. If there is a conflict in the voting results, the meta-model arbitration or rule base verification process will be triggered to generate the final analysis conclusion.

[0029] To facilitate understanding of the above technical solutions of the present invention, the following detailed description of the above technical solutions of the present invention is provided through specific embodiments.

[0030] Example 1 This invention provides a code intelligent analysis and auditing system based on a large language model. The system adopts a layered architecture, mainly including a front-end presentation layer, a core business processing layer, and a data persistence layer, and may further include an API gateway layer and an external service integration layer to enhance its functionality.

[0031] The front-end presentation layer provides a user interface and includes at least a project management module, an audit task module, and a code analysis module. Specifically, the project management module is responsible for project creation, list display, and detail management; the audit task module provides task creation, status monitoring, and result viewing functions; and the code analysis module integrates a real-time code analysis component and a file selector component, supporting users to perform real-time code snippet analysis or select files from the project for analysis, and visually displaying the analysis results through a visualization component. Furthermore, the front-end presentation layer may also include a configuration management module, a report generation module, and a system management module, respectively handling user configuration, report export, and system monitoring tasks.

[0032] The core business processing layer communicates with the front-end presentation layer and is the core of intelligent auditing. This layer includes a multi-source project acquirer, an intelligent code auditing engine, a multi-model collaborative scheduling and fusion engine, and a task scheduler.

[0033] The multi-source project acquirer supports importing code projects from multiple data sources. For example, it can acquire projects from remote repositories via an integrated GitHub API processor or GitLab API processor; or decompress and extract user-uploaded local compressed files via a ZIP file processor.

[0034] The intelligent code auditing engine is responsible for performing multi-dimensional analysis on the imported code, which includes at least security vulnerability detection, performance problem localization, and code maintainability assessment.

[0035] The multi-model collaborative scheduling and fusion engine is a key innovative component of this system. This engine schedules multiple heterogeneous large language models to perform parallel analysis on the same code unit, and fuses and arbitrates the outputs of each model to ultimately generate a unified and authoritative analytical conclusion. Its workflow includes: a parallel scheduling module concurrently requests services from multiple preset large language models; a result aggregation and arbitration module performs structured parsing of the results returned by each model, and performs weighted voting and semantic fusion based on preset model weights, historical accuracy, and confidence levels; in cases of conflicting conclusions, a meta-model arbitration or rule base verification process is triggered.

[0036] The task scheduler is responsible for auditing the creation, distribution, status tracking, and lifecycle management of tasks. It can achieve asynchronous execution and concurrency control of tasks through technologies such as Redis queues.

[0037] The data persistence layer stores all data during system operation. It includes at least an audit issue table for storing specific problems discovered during code analysis. Furthermore, this layer constructs an audit knowledge graph library to support the system's self-evolution and incremental analysis. The audit knowledge graph library may include an issue pattern table, a case sample table, a feedback learning record table, and a project audit history snapshot table. The issue pattern table records the characteristic fingerprints of various code issues, general remediation templates, and their severity rules. The case sample table stores typical positive and negative samples confirmed by humans, serving as a training dataset for model optimization. The feedback learning record table records human feedback on the results of artificial intelligence analysis and the dynamic adjustments made by the system accordingly, forming a traceable learning chain. The project audit history snapshot table stores the code state hashes and corresponding audit result snapshots of the project at the completion of each audit, serving as a comparison benchmark for performing incremental analysis.

[0038] Furthermore, the core business processing layer may also include a context-aware prompting engineering service and a dynamic software bill of materials analysis engine. The dynamic software bill of materials analysis engine parses project dependency files, generates a software bill of materials conforming to a standard format, and matches it with external vulnerability intelligence sources to construct a project security dependency graph. The context-aware prompting engineering service is responsible for integrating code context, security dependency graph, and project historical issues before scheduling the large language model, constructing enhanced prompts, and dynamically injecting associated vulnerability intelligence to improve analysis accuracy.

[0039] The system may also include an application programming interface (API) gateway layer positioned between the front-end presentation layer and the core business processing layer, providing standardized representational state transition API routing, request processing, and asynchronous communication mechanisms. Simultaneously, the system can integrate with external services such as GitHub and GitLab through an external service integration layer.

[0040] Example 2 This embodiment describes in detail the implementation steps of a code intelligent analysis and auditing method based on the above system.

[0041] First, users access the system's web interface through a browser to register and log in. Next, users create a new project, choosing between entering a Git repository's Uniform Resource Locator (URL) or uploading a ZIP compressed file. The system then invokes the multi-source project acquirer to clone the repository or decompress the file and initialize the project. Then, for an existing project, users configure parameters such as scan scope and exclusion modes to create a new audit task. The task scheduler receives and schedules this task.

[0042] Subsequently, the system performs intelligent code analysis. For the code files in the task, the intelligent code auditing engine drives the analysis process. During this process, the multi-model collaborative scheduling and fusion engine schedules multiple large language models to perform parallel analysis on the same code unit, and performs fusion arbitration on the results to generate a unified conclusion. The analysis dimensions include security, performance, code style, and maintainability. The result parser performs structured processing on the unified conclusions generated by the analysis, generating structured data including a list of issues, quality scores, and remediation suggestions.

[0043] The system stores the structured analysis results in a data persistence layer, including an audit issue table and related knowledge graph tables, and establishes relationships between projects, tasks, and issues. Based on the stored results, the reporting service automatically generates professional audit reports containing a project overview, issue details, and improvement suggestions, and supports export to multiple formats, including portable document formats.

[0044] Furthermore, the system drives an intelligent dual-loop process of incremental analysis and knowledge evolution through a task scheduler. In the incremental analysis loop, when project code changes, the system automatically triggers an incremental scan. By comparing the code status hash stored in the project audit history snapshot table, it accurately identifies the changed parts and their impact scope, analyzing only the incremental code to improve audit efficiency. In the knowledge evolution loop, the results of each audit, including confirmed issues, false alarm samples, and manual feedback, are structured and stored in the audit knowledge graph library. The system uses this data to continuously extract and optimize problem pattern features and dynamically adjusts the prompts submitted to the large language model, injecting historical context and verified rules, thereby enabling the system's audit judgment capabilities to continuously evolve.

[0045] In addition, the system provides an online interactive Q&A function, allowing users to ask further questions about the problems they discover, and the system provides intelligent answers through a large language model.

[0046] Example 3 This embodiment provides examples of key configurable parameters for system operation. For instance, the maximum concurrent scan count can be set to 10, the minimum detection interval to 30 minutes, and the default scan depth to 5 layers. The code analysis timeout can be set to 30 seconds, and the issue detection timeout rate is 20%. The code quality score threshold can be 80 points, and the security vulnerability score threshold can be 90 points. The report generation format supports portable document format, JSON, and CSV. The database type can use MySQL 8.0, and the caching database type can use Redis 6.2. The large model selection strategy can be dynamically selected based on project type and code volume. The model training cycle can be 7 days, the model training sample size can reach 1 million lines of code, and model evaluation metrics can include accuracy, recall, and F1 score.

[0047] For model optimization strategies, a continuous evolution mechanism based on a dynamic software bill of materials (BOM) and context awareness can be adopted. This mechanism includes context enhancement strategies and model fine-tuning and hint engineering strategies. The context enhancement strategy involves automatically scanning project dependencies using a software BOM analysis engine before code analysis, generating a dynamic software BOM, and constructing a security dependency graph by associating it with an external vulnerability database. During code auditing, the current code segment context is linked to this graph in real time, and relevant vulnerability information is dynamically injected into the hint words submitted to the large language model. The model fine-tuning and hint engineering strategy utilizes positive and negative samples accumulated in the audit knowledge graph library, especially those related to third-party vulnerabilities, to periodically fine-tune the core large language model, enhancing its ability to identify insecure dependency call patterns and achieving auditing across the code syntax layer and the software supply chain context layer.

[0048] System log levels can be set to information, warning, error, debug, etc. System monitoring metrics include CPU utilization, memory utilization, network connections, and task queue length. System alarm thresholds can be set as follows: CPU utilization ≥ 80%, memory utilization ≥ 90%, network connections ≥ 1000, and task queue length ≥ 1000.

[0049] In summary, by leveraging the technical solutions described above in this invention, and through the introduction of a multi-model collaborative scheduling and fusion mechanism, combined with context-aware prompting engineering and dynamic security dependency graphs, the system's understanding and analysis accuracy of deep code semantics and project-specific contexts is significantly improved. This effectively enhances the detection rate and accuracy of code security vulnerabilities and quality defects, while reducing false positives. Furthermore, by constructing an evolvable audit knowledge base and intelligent task scheduling, continuous optimization of analytical capabilities and a significant improvement in audit efficiency are achieved, ultimately realizing the integrated intelligent audit goal of supporting multi-source projects and providing professional reports.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A code intelligent analysis and auditing system based on a large model, characterized in that: include: The front-end presentation layer is used to provide the user interface and includes at least a project management module, an audit task module, and a code analysis module. The core business processing layer, which communicates with the front-end presentation layer, includes: A multi-source project acquirer to support importing code projects from remote repositories or local compressed files; The intelligent code auditing engine is used to perform multi-dimensional analysis on imported code projects. The multi-dimensional analysis includes at least security vulnerability detection, performance analysis, and maintainability assessment. A multi-model collaborative scheduling and fusion engine is used to schedule multiple heterogeneous large language models to perform parallel analysis on the same code unit, and to fuse and arbitrate the output results of each model to generate a unified analysis conclusion. The task scheduler is used to manage the execution of audit tasks; The data persistence layer is used to store system data, including at least an audit issue table for storing issues discovered during code analysis.

2. The code intelligent analysis and auditing system based on a large model according to claim 1, characterized in that, The multi-source project acquirer includes: GitHub API processors and / or GitLab API processors are used to retrieve project data from the corresponding remote repositories; A ZIP file processor used to decompress local compressed files and extract project data.

3. The code intelligent analysis and auditing system based on a large model according to claim 1, characterized in that, The multi-model collaborative scheduling and fusion engine includes: The parallel scheduling module is used to concurrently submit analysis requests for the same code unit to multiple preset large language model services; The results aggregation and arbitration module is used to perform structured parsing of the raw analysis results returned by each model, and to perform weighted voting and semantic fusion based on preset weights and confidence levels. For conflicting analysis results, meta-model arbitration or rule base verification is triggered to generate a final conclusion.

4. The code intelligent analysis and auditing system based on a large model according to claim 1, characterized in that, The core business processing layer also includes context-aware prompting engineering services and a dynamic software bill of materials analysis engine; The dynamic software bill of materials analysis engine is used to parse project dependency files, generate a software bill of materials in a standard format, and match it with external vulnerability intelligence sources to construct a project security dependency graph. The context-aware prompting engineering service is used to integrate code context, the security dependency graph, and project historical issue information before the multi-model collaborative scheduling and fusion engine scheduling of the large language model, to construct enhanced prompt words, and dynamically inject the associated vulnerability information in the security dependency graph into the enhanced prompt words.

5. The code intelligent analysis and auditing system based on a large model according to claim 1, characterized in that, The data persistence layer also includes an audit knowledge graph library, which includes: Issue pattern table, used to store the characteristic fingerprints of code issues and fix templates; The case sample table is used to store positive and negative samples that have been manually verified. The feedback and learning record form is used to record human feedback on the analysis results and the adjustments made by the system accordingly. The project audit history snapshot table is used to store the code status hash and audit result snapshots at the time of each audit to support incremental analysis.

6. The code intelligent analysis and auditing system based on a large model according to claim 1, characterized in that, The task scheduler is also used to drive the incremental-knowledge dual-loop audit process, including: The incremental analysis loop is used to trigger an incremental scan by comparing the project audit history snapshot table after code changes. The knowledge evolution loop is used to structurally store each audit result in the audit knowledge graph library, and dynamically optimize the prompt words submitted to the large language model based on the accumulated case samples and problem patterns.

7. The code intelligent analysis and auditing system based on a large model according to claim 1, characterized in that, The code analysis module in the front-end presentation layer includes: The real-time code analysis component is used to receive code snippets input by the user and trigger real-time analysis; The file selector component is used to browse and select files to be analyzed within the project file; An analysis results visualization component used to display code analysis results.

8. The code intelligent analysis and auditing system based on a large model according to claim 1, characterized in that, The system also includes: The API gateway layer is set between the front-end presentation layer and the core business processing layer to provide standardized application programming interface routing and request processing. The external service integration layer is used to integrate and provide a unified interface for interacting with GitHub and GitLab services.

9. A code intelligent analysis and auditing method based on a large model, applied to the system as described in any one of claims 1-8, characterized in that, Including the following steps: Import code projects using the multi-source project acquirer; Create an audit task for the code project using the task scheduler; The intelligent code auditing engine performs multi-dimensional analysis of the code files in the code project; During the analysis process, multiple heterogeneous large language models are scheduled in parallel to analyze the same code unit through multi-model collaborative scheduling and fusion engine scheduling, and the analysis results are fused and arbitrated to generate a unified conclusion. Issues identified during the analysis are stored in the audit issue table of the data persistence layer.

10. The code intelligent analysis and auditing method based on a large model according to claim 9, characterized in that, The steps of performing parallel analysis of the same code unit by scheduling multiple heterogeneous large language models through multi-model collaborative scheduling and fusion engine scheduling, and then fusion and arbitrating the analysis results to generate a unified conclusion include: Concurrently submit analysis requests to multiple large language model services; Receive the analysis results returned by each model and perform structured parsing; The parsed results are weighted and fused semantically based on preset model weights, historical accuracy, and real-time confidence. If there is a conflict in the voting results, the meta-model arbitration or rule base verification process will be triggered to generate the final analysis conclusion.

Images