RAG (Retrieval-Augmented Generation) Project

This project implements a Retrieval-Augmented Generation (RAG) system using LangChain, MongoDB Atlas Vector Search, and Ollama models. The system is capable of processing various document types and answering questions based on the content.

Project Structure

The project consists of two main components:

1. script.py: A production-ready implementation using LangChain and MongoDB Atlas
2. project.ipynb: A Jupyter notebook implementation using Llama.cpp and FAISS

script.py

This script implements a RAG system with the following features:

Key Components

• Document Processing:

  • Supports multiple document formats: PDF, CSV, DOCX, and HTML
  • Uses LangChain's document loaders for parsing
  • Implements text chunking with RecursiveCharacterTextSplitter

• Vector Storage:

  • Uses MongoDB Atlas Vector Search for document storage and retrieval
  • Implements efficient vector search with configurable parameters
  • Supports async operations for better performance

• Language Models:

  • Uses Ollama for both embeddings and text generation
  • Implements a structured prompt template for consistent responses
  • Supports temperature control for response generation

Usage

1. Set up environment variables in .env:

   MONGO_URI=your_mongodb_connection_string
   pdf_path=path_to_pdf
   csv_path=path_to_csv
   docx_path=path_to_docx
   html_path=path_to_html

2. Run the script:

   python script.py

project.ipynb

This Jupyter notebook implements a simpler RAG system using different technologies:

Key Components

• Document Processing:

  • Uses simple text chunking (500 characters per chunk)
  • Implements basic text file reading

• Vector Storage:

  • Uses FAISS for efficient vector similarity search
  • Implements a simple chunk mapping system

• Language Models:

  • Uses Llama.cpp for text generation
  • Uses SentenceTransformer for embeddings
  • Implements a simple question-answering system

Usage

1. Install required packages:

   pip install llama-cpp-python faiss-cpu sentence-transformers

2. Prepare your model:

   • Download the Llama 2 model (llama-2-7b-chat.Q4_K_M.gguf)
   • Place it in the project directory

3. Run the notebook:

   • Open project.ipynb in Jupyter
   • Execute cells in sequence
   • Use the interactive question-answering interface

Key Differences

• script.py is more production-ready with:

  • Support for multiple document types
  • MongoDB Atlas integration
  • Async operations
  • More robust error handling

• project.ipynb is more experimental with:

  • Simpler implementation
  • Local vector storage (FAISS)
  • Interactive notebook interface
  • Easier to modify and experiment with

Requirements

All required packages are listed in requirements.txt. You can install them using:

pip install -r requirements.txt

The main dependencies include:

Core Dependencies

• python-dotenv - For environment variable management
• pymongo - For MongoDB integration

LangChain Dependencies

• langchain - Core LangChain functionality
• langchain-ollama - Ollama integration
• langchain-mongodb - MongoDB vector store integration
• langchain-community - Community integrations

Document Processing

• PyPDF2 - PDF processing
• python-docx - DOCX processing
• unstructured - HTML and other document processing
• docx2txt - Additional DOCX support

Vector Storage and Embeddings

• faiss-cpu - For local vector similarity search
• sentence-transformers - For text embeddings
• numpy - Required for numerical operations

LLM Dependencies

• llama-cpp-python - For local LLM inference
• torch - Required for transformers
• transformers - For model loading and inference

Setup

1. Clone the repository
2. Install dependencies:

   pip install -r requirements.txt

3. Set up environment variables in .env
4. Download required models:
   • For script.py: No additional model downloads needed (uses Ollama)
   • For project.ipynb: Download llama-2-7b-chat.Q4_K_M.gguf
5. Run either the script or notebook based on your needs