https://github.com/kivanc57/react_rag_llama_crew

React&RAG Llama Crew is an AI-powered system using LlamaIndex, RAG, and Hugging Face embeddings for code retrieval, analysis, and generation. A multi-agent crew—Llama3.2, CodeLlama, Ollama, and ReActAgent—handles queries, documentation, and structured code generation. 🚀
https://github.com/kivanc57/react_rag_llama_crew

agent-crew ai code-llama gen-ai huggingface-transformers llama llama-index llm ollama rag react-agent

Last synced: 6 months ago
JSON representation

React&RAG Llama Crew is an AI-powered system using LlamaIndex, RAG, and Hugging Face embeddings for code retrieval, analysis, and generation. A multi-agent crew—Llama3.2, CodeLlama, Ollama, and ReActAgent—handles queries, documentation, and structured code generation. 🚀

• Host: GitHub
• URL: https://github.com/kivanc57/react_rag_llama_crew
• Owner: kivanc57
• License: gpl-2.0
• Created: 2025-02-09T17:23:04.000Z (8 months ago)
• Default Branch: master
• Last Pushed: 2025-02-09T17:35:26.000Z (8 months ago)
• Last Synced: 2025-02-09T18:31:50.541Z (8 months ago)
• Topics: agent-crew, ai, code-llama, gen-ai, huggingface-transformers, llama, llama-index, llm, ollama, rag, react-agent
• Language: Python
• Homepage:
• Size: 0 Bytes
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# React&RAG LLama Crew

## 🚀 Overview

This project introduces an **AI Agent Crew** powered by **LlamaIndex** to analyze, generate, and process code efficiently. The system consists of **multiple specialized LLMs and a ReActAgent**, forming a collaborative AI team that enhances Retrieval-Augmented Generation (RAG), structured output parsing, and vector-based search.

## 🔥 Key Features

### 🏆 Multi-Agent AI Crew

- **Llama3.2**: Handles natural language understanding and analysis.

- **CodeLlama**: Specialized in code generation and improvement.

- **Ollama**: Supports enhanced query response and knowledge retrieval.

- **ReActAgent**: Coordinates tools, interacts with the user, and ensures logical reasoning.

### 🧠 Intelligent Code Analysis & Generation

- Generates, refines, and explains code based on user prompts.

- Understands complex queries and adapts responses accordingly.

### 📄 API Documentation Querying

- **Vector-based search** on API documentation for instant lookups.

- **Automatic document processing** for `.pdf` files.

- **LlamaParse integration** to extract meaningful text from documents.

### 🔎 Retrieval-Augmented Generation (RAG)

- **Combines LLMs with real-time document retrieval** to enhance code generation accuracy.

- **Ensures responses are grounded in relevant documentation**.

- Reduces hallucinations and improves factual correctness.

### 📂 Code Reader Functionality

- Reads and retrieves code files using the **code_reader** tool.

- Helps in debugging, refactoring, and understanding existing code.

### 🔗 LlamaIndex-Powered Query Pipeline

- **VectorStoreIndex** for efficient document retrieval.

- **QueryEngineTool** for answering API documentation queries.

- **SimpleDirectoryReader** to load and process data.

### ⚙️ Advanced Code Parsing & Output Structuring

- Uses **PydanticOutputParser** to structure output in JSON format.

- Generates **clean, structured** code with meaningful descriptions and filenames.

### 🎯 Robust & Reliable Execution

- **Error handling & retry mechanism** for resilient processing.

- Saves generated code **automatically** to the output directory.

## 🛠 How It Works

1. **User Input**: Enter a prompt to generate or analyze code.

2. **Processing**: The agent crew retrieves relevant documentation, reads code, and formulates responses.

3. **Generation**: The AI produces well-structured code and descriptions.

4. **Output**: The code is saved with an appropriate filename.

## 🏗 Technologies Used

- **LlamaIndex**: Efficient document indexing and retrieval.

- **Ollama**: Llama3.2 & CodeLlama models for text & code processing.

- **Pydantic**: Structured output parsing.

- **HuggingFace Optimum**: High-performance vector embeddings.

- **Retrieval-Augmented Generation (RAG)**: Enhances AI responses with real-time document retrieval.

- **Ast & Dotenv**: Code execution & environment configuration.

## 📌 Usage

1. Place code and documentation files in the `data/` folder.

2. Run the script:  

   ```bash

   python main.py

   ```

3. Follow the prompt to analyze or generate code.

4. Output files are saved in the `output/` folder.

## 📩 Contact

For improvements or issues, feel free to contribute or reach out!