🔍 RAG with MongoDB

RAG (Retrieval-Augmented Generation) application using MongoDB Atlas Vector Search to enhance LLM-powered applications through semantic information retrieval.

• Batch PDF document processing
• Automatic metadata extraction using LLM
• Vector embeddings with OpenAI
• MongoDB Atlas Vector Search integration
• Semantic search capabilities

This project is part of MongoDB University educational materials. Part of MongoDB University - MongoDB Skills course: RAG with MongoDB

⚙️ MongoDB Atlas Setup

1. Create Database and Collection

• Database: documents
• Collection: docs-chunks

2. Create Vector Search Index

Navigate to Atlas Search and create a new Vector Search Index:

Index Name: vector_index

JSON Configuration:

{
  "fields": [
    {
      "type": "vector",
      "path": "embedding",
      "numDimensions": 1536,
      "similarity": "cosine"
    },
    {
      "type": "filter",
      "path": "hasCode"
    }
  ]
}

👨🏻‍💻 How to use

• Prerequisites:
  • Python 3.13+
  • MongoDB Atlas account with Vector Search Index configured
  • OpenAI API key

1. Install dependencies:

pip install langchain langchain_community langchain_core langchain_openai langchain_mongodb pymongo pypdf python-dotenv

2. Create a .env file in the root directory:

OPENAI_API_KEY=your-openai-api-key
MONGODB_URI=your-mongodb-connection-string

3. Place your PDF files in the docs/ folder

4. Run the ingestion script:

python load_data.py

The script will automatically process all PDFs in the docs/ folder and:

• Load and clean each PDF
• Extract metadata (title, keywords, code detection)
• Split documents into chunks (500 chars with 150 overlap)
• Generate embeddings using OpenAI
• Store everything in MongoDB Atlas

5. After ingesting documents, you can query your data:

python rag.py

Or import and use programmatically:

from rag import query_data

answer = query_data("What is the difference between a collection and database in MongoDB?")
print(answer)