LLM Engineering Platform

简体中文 | English

An intelligent open-source project assessment platform based on LLMs and Multi-Agent Systems.

Project Overview

This system utilizes Large Language Models (LLMs) and a dynamic Multi-Agent System (MAS) architecture to provide comprehensive evaluations of open-source projects. Key features include:

• User intent recognition and parsing.
• Dynamic task decomposition and collaborative execution by specialized agents.
• Agent registration, discovery, and A2A/MCP communication.
• Sophisticated prompt engineering using Jinja2 templates.
• Integration with diverse knowledge sources:
  • Version Control: GitHub, Gitee APIs.
  • Vulnerability Databases: OSV.
  • Graph Databases: Neo4j for relationship analysis.
  • Vector Stores: FAISS for semantic search.
  • Web Search: Real-time information retrieval.
• Extensible evaluation metrics.
• Structured report generation with visualizations.

Setup and Installation

1. Clone this repository.
2. Create a virtual environment: python -m venv venv
3. Activate the virtual environment:
   • Linux/macOS: source venv/bin/activate
   • Windows: venv\Scripts\activate
4. Install dependencies: pip install -r requirements.txt
5. Configure API keys and other settings in app/config.py or via environment variables (e.g., a .env file).
6. Run the application: uvicorn app.main:app --reload

Project Structure

(Please refer to the generated file structure and comments within the files for details.)

Contributing

Contributions are welcome! Please refer to CONTRIBUTING.md for guidelines.

License

This project is licensed under the MIT License.