GraphRAG Implementation for Healthcare Product Catalog

This repository contains a GraphRAG (Graph-enhanced Retrieval-Augmented Generation) implementation for a healthcare company's product catalog using the official Microsoft GraphRAG package. The implementation automatically extracts entities and relationships from unstructured text documents to build a knowledge graph, which is then used to enhance retrieval and answer questions.

What is GraphRAG?

GraphRAG is an approach that combines the strengths of knowledge graphs with retrieval-augmented generation. It addresses limitations of traditional RAG systems by:

1. Automatically extracting entities and relationships from documents
2. Building a knowledge graph to represent structured information
3. Using the graph structure to enhance retrieval beyond simple vector similarity
4. Integrating graph-based and vector-based retrieval for more comprehensive answers

How It Works

This implementation uses the official Microsoft GraphRAG CLI to:

1. Index documents: The system processes text files in the input directory, extracting entities and relationships to build a knowledge graph.
2. Query the graph: The system supports both global and local search methods to answer questions about the healthcare products.
3. Create statistics: The system generates statistics about the knowledge graph, such as the number of entities and relationships.
4. Visualize the knowledge graph: The system generates a visual representation of the entities and relationships in the knowledge graph.

Project Structure

• pipeline.py: The main implementation file that contains the GraphRAGPipeline class for indexing and querying
• analyzer.py: Contains the GraphRAGAnalyzer class for analyzing and visualizing the knowledge graph
• main.py: Example script demonstrating the GraphRAG functionality
• input/: Directory containing the input text files
• output/: Directory where GraphRAG stores its output files (entities, relationships, etc., will be created by GraphRAG)
• logs/: Directory for log files (will be created by GraphRAG)
• cache/: Directory for cached data (will be created by GraphRAG)

Requirements

• Python 3.12 (managed by uv)
• An OpenAI API key

Installation

1. Install uv if you don't have it (install guide)

2. Install dependencies — uv reads pyproject.toml and uv.lock and provisions Python 3.12 if needed

uv sync

Project Setup

1. Scaffold the graphrag config (creates settings.yaml and prompts/). Accept the defaults — the 3.x templates use gpt-4.1 and text-embedding-3-large.

uv run graphrag init --root ./

Warning: graphrag init will overwrite an existing .env in the project root with a placeholder. If you already have one, back it up first (cp .env .env.bak).

2. Create .env with your OpenAI key

echo 'GRAPHRAG_API_KEY=<API_KEY>' > .env

3. Approve environment settings

direnv allow

Usage

Run the example script:

uv run python main.py

The script will:

• Run the indexing process
• Execute example search queries
• Generate statistics and visualization of the knowledge graph

Features

• CLI-based interaction: Uses the GraphRAG CLI for indexing and querying
• Graph analysis: Provides statistics and insights about the knowledge graph
• Knowledge graph visualization: Creates visual representations of entities and relationships

References

• Microsoft GraphRAG GitHub Repository
• Microsoft Research Blog Post
• Blog Post