Docling RAG System

A powerful Retrieval-Augmented Generation (RAG) system built with Python that leverages document processing, vector-based search, and LLM integration to provide accurate answers based on your document collection.

🌟 Features

• Powerful Document Processing: Automatic processing of various document formats including PDF, DOCX, HTML, TXT and MD
• Semantic Search: Find relevant information using meaning-based search powered by embeddings
• Hybrid Search: Combine semantic and lexical search for improved accuracy
• Context-Aware Answers: Generate answers to questions based on document context
• Flexible API: Use via command line or REST API
• Customizable: Configurable components and parameters for different use cases

📋 Quick Start

Prerequisites

• Python 3.8+
• [Optional] GPU with CUDA support for faster processing

Installation

# Clone the repository
git clone https://github.com/yourusername/docling-rag.git
cd docling-rag

# Set up virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt
# install spacy lib needed
python -m spacy download fr_core_news_lg

Basic Usage

1. Process your documents:

   python main.py process --dir your_documents_directory

2. Index the processed documents:

   python main.py index

3. Search for information:

   python main.py search "your search query"

4. Ask questions about your documents:

   python main.py ask "your question"

5. Start the API server:

   python server.py

📚 Documentation

For more detailed documentation, see:

• Installation Guide
• Code Structure
• API Reference
• Configuration Guide
• Contribution Guidelines

🧪 Example

# Python API usage example
from rag_orchestrator_api import RAGOrchestratorAPI
import asyncio

async def main():
    # Create orchestrator
    rag = RAGOrchestratorAPI()
    
    # Process a document
    result = await rag.process_document("path/to/document.pdf")
    print(f"Document processed: {result['document_id']}")
    
    # Index documents
    await rag.index_documents()
    
    # Search for information
    search_results = await rag.search("vector databases")
    
    # Ask a question
    answer = await rag.ask("How do vector databases work?")
    print(f"Answer: {answer['answer']}")

if __name__ == "__main__":
    asyncio.run(main())

🔧 Configuration

Configuration is managed through config.py and config.json. Key settings include:

• Data Paths: Configure where documents and processed data are stored
• Embedding Model: Select the embedding model for vectorization
• Vector Database: Choose between FAISS or Chroma
• Chunking Parameters: Configure how documents are split into chunks
• Search Settings: Adjust search parameters for optimal results

🚀 Advanced Features

• Cross-Encoder Reranking: Improve search result ranking
• Hybrid Search: Combine semantic and lexical search capabilities
• Document Enrichment: Extract code, formulas, and images from documents
• Custom Chunking: Semantic-based document splitting for better context

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgements

• Docling for document processing components
• LangChain for vector store and embeddings integrations
• HuggingFace for transformer models
• FastAPI for the REST API framework