RAG (Retrieval-Augmented Generation) Knowledge Base

This project implements a Retrieval-Augmented Generation (RAG) system that allows you to process documents, store them in a vector database, and query them using natural language. The system uses Qdrant for vector storage, Hugging Face's RoSBERTa for embeddings, and a local GGUF model (T-lite-it-1.0) for generating responses. It also features LangGraph workflows for structured RAG pipelines and LangSmith tracing for monitoring and debugging.

Features

• Document processing for various file formats (PDF, DOCX, HTML, plain text)
• Vector embeddings using Hugging Face's RoSBERTa model
• Vector storage and retrieval with Qdrant
• Interactive chat interface with Gradio
• Local LLM inference with llama-cpp-python
• Hybrid search combining semantic and keyword-based retrieval
• LangGraph integration for structured RAG workflows
• LangSmith tracing for monitoring and debugging performance
• Support for multiple document formats in vector storage (flexible field mapping)

Prerequisites

• Python 3.8+
• Qdrant vector database (running on localhost:6333)
• GGUF Model (T-lite-it-1.0-Q4_K_M-GGUF included)
• Poetry (recommended) or pip
• LangChain and LangGraph (included in requirements.txt)
• LangSmith API key (optional, for tracing and monitoring)

Installation

1. Clone the repository:

   git clone <your-repository-url>
   cd rag

2. Create and activate a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   # For Ubuntu/Debian:
   sudo apt-get update
   sudo apt-get install -y build-essential cmake python3-dev python3-venv
   
   # For CentOS/RHEL:
   sudo yum groupinstall -y "Development Tools"
   sudo yum install -y cmake python3-devel
   
   # Install Python dependencies
   pip install -r requirements.txt

4. Download and run Qdrant (using Docker):

   docker pull qdrant/qdrant
   docker run -p 6333:6333 -p 6334:6334 qdrant/qdrant

5. Download the GGUF model (if not included):

   • The project includes a pre-configured model in model/T-lite-it-1.0-Q4_K_M-GGUF/
   • To use a different model, update the path in gradio_app.py

Project Structure

• data/ - Directory for storing documents to be processed
• model/ - Contains the GGUF model files
• data_processing.py - Script for processing and indexing documents
• rag_app.py - Main RAG application with LangGraph workflows and Gradio interface
• prompts.py - System prompts and templates
• requirements.txt - Python dependencies

Usage

1. Prepare Your Documents

Place the documents you want to process in the data/ directory. Supported formats include:

• Text files (.txt)
• PDF documents (.pdf)
• Word documents (.docx)
• HTML files (.html, .htm)

2. Process Documents

Run the document processing script to create vector embeddings and store them in Qdrant:

python data_processing.py

This will:

1. Load documents from the data/ directory
2. Split them into chunks
3. Generate embeddings using RoSBERTa
4. Store them in the Qdrant database

3. Start the Chat Interface

Start the Gradio chat interface:

python rag_app.py

This will start a local web server (usually at http://localhost:7860) with the chat interface where you can ask questions about your documents.

Configuration

You can customize the following aspects of the system:

• Model Parameters: Adjust n_ctx, n_threads, and other parameters in the LangChainAssistant class
• Search Settings: Modify the hybrid search parameters in HybridSearch class
• UI Settings: Customize the Gradio interface in create_demo()
• LangGraph Workflow: Modify the RAG workflow in RAGAssistant._create_rag_graph()
• LangSmith Tracing: Configure with environment variables LANGCHAIN_API_KEY and LANGCHAIN_PROJECT
• Chunk size and overlap: Adjust in data_processing.py
• Qdrant connection: Modify the connection parameters in Config class

Troubleshooting

Common Issues

1. Qdrant connection issues:

   • Make sure Qdrant is running (docker ps should show the container)
   • Check if ports 6333 and 6334 are available

2. OCR issues:

   • Verify Tesseract is installed and in your PATH
   • For non-English text, you might need to install additional language packs

3. Vector dimension mismatch:

   • If you see "Vector dimension error" from Qdrant, check the embedding dimensions in SimpleEmbeddings
   • The system now automatically pads vectors to the expected 1024 dimensions

4. LangGraph and LangSmith integration:

   • Do not use @trace decorators on functions passed to LangGraph's add_node
   • Instead, use with trace(name="function_name"): inside functions
   • This prevents the "trace object not callable" error

Memory Management

For large document collections, you might need to increase the available memory for Python and Ollama.

License

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

Acknowledgements

• LangChain
• LangGraph
• LangSmith
• Qdrant
• Hugging Face

---

For any questions or issues, please open an issue in the repository.