Skip to content

README.md

Latest commit

 

History

History
133 lines (95 loc) · 3.87 KB

README.md

File metadata and controls

133 lines (95 loc) · 3.87 KB

RAG Chat API Starter Kit

A RAG (Retrieval Augmented Generation) system that uses Google's Gemini models to provide contextual responses based on your documentation.

Overview

RAG Chat API enables you to create a chatbot that answers questions based on your technical documentation. It uses:

  • Google Gemini models for advanced embeddings and text generation
  • ChromaDB for vector storage and similarity search
  • LangChain for document loading, chunking, and prompt management

This system creates embeddings of your documentation, stores them in a vector database, and then retrieves relevant context when answering user queries.

Features

  • ✅ Load and process documents (.txt, .md, .pdf (expected to be introduced in v2.0.0))
  • ✅ Split documents into manageable chunks with overlap
  • ✅ Generate and store embeddings with Google's text-embedding-004 model
  • ✅ Semantic search for relevant document chunks
  • ✅ Contextual answers using Gemini 1.5 Pro
  • ✅ Customizable chunk size and overlap
  • ✅ Organized response storage

Installation

  1. Clone this repository:

    git clone <repository-url>
cd devdocs-chat-api

  2. Create and activate a virtual environment:

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

  3. Install dependencies:

    pip install -r requirements.txt

  4. Create a .env file with your configuration:

    CHROMA_PATH="db/chroma"
GOOGLE_API_KEY="your-google-api-key-here"
DATA_STORE_PATH="db/knowledge_base"

Usage

1. Add your documentation

Place your documentation files in the directory specified by DATA_STORE_PATH (default: knowledge_base):

mkdir -p db/knowledge_base
cp your-docs/*.md db/knowledge_base/

2. Generate embeddings

Run the embeddings script to process your documentation and create vector embeddings:

python embeddings.py

This will:

  • Load all documents from your knowledge base directory
  • Split them into chunks of text
  • Generate embeddings using Google's text-embedding-004 model
  • Store the embeddings in ChromaDB

3. Query your documentation

Run the main script to interact with your documentation:

python main.py

By default, this will:

  • Ensure your knowledge base and ChromaDB directories exist
  • Check if there are documents to process
  • Generate embeddings if needed
  • Run a sample query ("Why do we need to use embeddings?")
  • Find relevant document chunks based on the query
  • Generate a contextual response using Gemini 1.5 Pro
  • Save the response to the responses directory

Customization

You can customize the behavior by modifying:

  • Query: Change the example query in main.py
  • Chunking parameters: Adjust CHUNK_SIZE and CHUNK_OVERLAP in embeddings.py
  • Model: Change the embedding or LLM model in the respective files
  • Prompt template: Modify the PROMPT_TEMPLATE in main.py

Project Structure

  • main.py: Main script for querying the documentation
  • embeddings.py: Script for generating and storing embeddings
  • .env: Environment variables
  • db: Directory for databases
    • chroma/: ChromaDB vector database
    • knowledge_base/: Your documentation files
  • responses: Generated responses

Requirements

  • Python 3.9+
  • Google API key with access to Gemini models
  • Dependencies listed in requirements.txt

Troubleshooting

  • Empty ChromaDB: Make sure you have documents in your knowledge base directory and run embeddings.py
  • No responses: Check that your Google API key is valid and has access to the required models
  • Missing directories: The system will create required directories automatically

License

[Your License Here]

Contributing

Contributions are welcome! Please fork the repository and submit a pull request with your changes. Make sure to follow the code style and include tests for new features.