GraphRAG-MedIntel

• 

  Flowchart

• 

  Demo Flow

• 

  graph

• 

  neo4j demo

🚀 Inspiration

Healthcare data is vast, scattered, and complex — from research papers to patient records and drug interactions. Traditional AI chatbots often hallucinate or miss critical links between diseases, drugs, and symptoms. We wanted to create a system that doesn’t just retrieve information, but understands relationships — bridging retrieval and reasoning for reliable medical insights.

💡 What it does

GraphRAG-MedIntel is an intelligent medical assistant that combines Retrieval-Augmented Generation (RAG) with Graph Database reasoning.

It retrieves medical information from research data using embeddings and FAISS.

It maps entities (like drugs, diseases, and symptoms) into a Neo4j knowledge graph.

It performs multi-hop reasoning through the graph to generate accurate, explainable answers.

Essentially, it helps users query complex relationships like:

“What are the interactions between Aspirin and Ibuprofen?” and gives not just the answer, but the reasoning path behind it.

🛠️ How we built it

Data Processing: Extracted structured triples (drug–disease–symptom relationships) from medical datasets.

Embeddings: Used SentenceTransformers for semantic similarity and FAISS for fast retrieval.

Graph Construction: Built a Neo4j graph to store entities and relationships.

Backend Logic: Developed a hybrid RAG pipeline combining vector retrieval with graph reasoning.

Frontend: Implemented a Streamlit UI for natural language queries and graph visualizations.

⚔️ Challenges we ran into

Integrating vector and graph retrieval seamlessly.

Dealing with entity duplication and normalization across datasets.

Optimizing Neo4j queries for multi-hop reasoning.

Balancing accuracy vs. speed in large retrieval tasks.

🏆 Accomplishments that we're proud of

Successfully created a working hybrid RAG system combining FAISS and Neo4j.

Achieved explainable outputs with reasoning paths visualized through the graph.

Designed a clean, interactive UI for intuitive medical data exploration.

📚 What we learned

How to build and query knowledge graphs with Neo4j.

How RAG architectures can be extended beyond simple retrieval.

The power of combining symbolic reasoning (graphs) with neural embeddings.

🔮 What's next for GraphRAG-MedIntel

Integrate real-time PubMed and FDA API data for live updates.

Implement LLM fine-tuning for domain-specific question answering.

Expand to other verticals like finance or legal knowledge graphs.

Deploy as a secure API for hospitals or researchers

Built With

• faiss
• langchain
• neo4j
• python
• pytorch