RAG with MongoDB Atlas & LlamaIndex

This repository provides a clean implementation of a Retrieval-Augmented Generation (RAG) system using MongoDB Atlas as a Vector Database and LlamaIndex for orchestration.

Features

• Vector Ingestion: Process local documents and store them as high-dimensional vectors in MongoDB.
• Semantic Retrieval: Uses MongoDB Atlas Vector Search to find relevant context.
• Contextual Generation: Powered by OpenAI to provide accurate answers based on your private data.

Project Structure

• load_data.py: Script to initialize the vector store and upload your document embeddings.
• rag.py: The query engine used to ask questions against the indexed data.
• key_param.py: Configuration file for API keys and connection strings.

Setup Instructions

1. Prerequisites

• Python 3.9+
• A MongoDB Atlas Cluster (M0 tier or higher).
• An OpenAI API Key.

2. Configure MongoDB Vector Search

You must create a Vector Search index in your MongoDB Atlas dashboard. Use the name vector_index and ensure the path is set to embedding with 1536 dimensions (for OpenAI embeddings) using cosine similarity.

3. Installation

Run the following commands to set up your environment:

• git clone https://github.com/muhammadcaeed/rag_with_mongodb.git
• cd rag_with_mongodb
• pip install llama-index pymongo llama-index-vector-stores-mongodb llama-index-embeddings-openai

4. Configuration

Update your key_param.py file with your specific credentials including your MONGO_URI, OPENAI_API_KEY, and the specific Database and Collection names you intend to use.

Usage

1. Load your data: Place your source documents in a folder named /data and run python load_data.py.
2. Query the system: Once indexed, run python rag.py to start the interactive query engine.