MedIntel+ — Graph RAG Health Copilot (Hackathon MVP)

What

MedIntel+ is a demo Graph RAG system combining Neo4j (knowledge graph), FAISS (semantic retrieval), and an LLM (OpenAI or Ollama) to answer medical knowledge queries with explainable evidence.

Quick setup (local)

1. Clone or copy the project to your machine and cd into it.

2. Create virtual env & install:

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

3. Run Neo4j (Docker recommended):

docker run -d --name neo4j -p7474:7474 -p7687:7687 -e NEO4J_AUTH=neo4j/test neo4j:latest
# open http://localhost:7474 and login neo4j/test

4. (Optional) create a .env file at repo root with:

NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASS=test
OPENAI_API_KEY=sk-...
LLM_BACKEND=openai   # or 'ollama'
OLLAMA_URL=http://localhost:11434
OLLAMA_MODEL=mistral

5. Build graph & vector store:

python graph/graph_build.py
python vectorstore/ingest_embeddings.py

6. Run UI:

streamlit run app/app.py

Deployment

1. Set up environment variables in your deployment environment:

   # Required
   NEO4J_URI=your-neo4j-uri
   NEO4J_USER=your-neo4j-user
   NEO4J_PASS=your-neo4j-password
   OPENAI_API_KEY=your-openai-key
   
   # Optional with defaults
   FAISS_INDEX_PATH=medintel_plus/vectorstore/med_faiss.index
   FACTS_PICKLE=medintel_plus/vectorstore/med_facts.pkl
   LLM_BACKEND=openai
   OLLAMA_URL=http://localhost:11434
   OLLAMA_MODEL=mistral

2. Initialize the vectorstore:

   python -m vectorstore.ingest_embeddings

   This will create the FAISS index and facts pickle file in the specified locations.

3. Initialize the knowledge graph:

   python -m graph.graph_build

   This will populate your Neo4j database with the medical knowledge graph.

4. Start the application:

   streamlit run app/app.py

Note: The app will automatically rebuild the vectorstore if the index files are missing.

Notes

• This is a demo; always recommend clinician verification.
• For offline demo, use Ollama and set LLM_BACKEND=ollama.

GraphRAGMed

MedIntel+ MedIntel+ is an AI-powered health copilot that answers medical questions using a combination of knowledge graph reasoning and retrieval-augmented generation (RAG). It leverages both semantic search (vector embeddings) and graph-based evidence to provide trustworthy, explainable answers for drugs, diseases, side effects, and interactions