Chat with ReadTheDocs using RAG

A Python application that enables conversational interaction with ReadTheDocs documentation using Retrieval-Augmented Generation (RAG). While demonstrated using Gramine SGX's documentation (included by default), the application can work with any ReadTheDocs content that has been locally scraped. The system features history-aware retrieval, allowing for natural follow-up questions and maintaining conversation context.

This project is part of the exercises for the LangChain- Develop LLM powered applications with LangChain course by Eden Marco, and is available via Udemy here. See Section 6 in the course for Marco's version of the project. The main differences are:

• I use Ollama instead of OpenAI for the embeddings
• I use Chroma instead of PineCone for the vector store
• I have ReadTheDocs of Gramine SGX instead of Langchain for illustration.

For anyone looking to learn LangChain, I highly recommend Marco's course!

Features

• ReadTheDocs documentation ingestion and embedding storage
• Context-aware chat interface that maintains conversation history
• Intelligent query reformulation using LangChain's rephrase prompts
• Source attribution for answers
• Batched document processing with progress tracking

Technologies Used

• Python 3.x
• LangChain
• OpenAI GPT models for query answering
• Ollama (local) for embeddings
• ChromaDB for vector storage
• Streamlit for the chat interface

Two versions of the chat interface:

• app.py: Basic chat functionality
• enhanced_app.py: Enhanced version with additional styling and UI improvements (created using Cursor AI's composer functionality)

Setup

1. Install the required dependencies:

pip install -r requirements.txt

2. Set up your OpenAI API key:

   • Create a .env file in the root directory
   • Add your OpenAI API key:

     OPENAI_API_KEY=your_api_key_here
     

3. Run either version of the chat interface from the python directory:

streamlit run app.py

or

streamlit run enhanced_app.py

4. (Optional) To chat with different ReadTheDocs documentation:
   • Scrape and store your chosen ReadTheDocs content locally
   • Update the documentation path in the configuration

License

This project is licensed under the Apache License, Version 2.0 (APL 2.0). License. See the LICENSE file for details.