maheshsmc/graph-rag-d11

Day 11 — Mini Graph‑RAG (Gradio + NetworkX)

A tiny, intern‑friendly Graph‑RAG demo you can deploy on Hugging Face Spaces.

✨ What it shows

• —How to store facts as a graph (nodes & edges).
• —How to answer questions via relation lookup + simple multi‑hop traversal.
• —A lightweight UI with Gradio.

🗂️ Files

• —app.py — Gradio UI.
• —graph_rag.py — Graph-RAG logic with a small demo knowledge graph.
• —requirements.txt — minimal deps.
• —docs/ — put a screenshot of the running app here (optional but recommended).

▶️ Run locally

pip install -r requirements.txt
python app.py

Then open the local URL shown by Gradio.

🚀 Deploy on Hugging Face Spaces

1. 1.Go to Hugging Face → New Space.
2. 2.Space SDK: Gradio. Hardware: CPU basic is enough.
3. 3.Upload these files (or drag‑drop the ZIP below).
4. 4.Wait for the build; then test with:
5. 5.Who introduced Obamacare?
6. 6.Who was president when Obamacare was introduced?

🧠 How it works (short)

• —We create a directed graph of triples (subject, relation, object) using NetworkX.
• —For “who introduced X”, we find edges with relation introduced that match X (with simple aliases).
• —For “who was president when X was introduced”, we do two hops:
• —introduced(X) → find the person.
• —Check if that person has an edge was_president_of → return them.

This is intentionally minimal. In a real app, replace the naive keyword checks with embeddings and add a proper graph DB (e.g., Neo4j).

📸 Screenshot

Add a screenshot of your Space UI to docs/ and commit, so readers see it in the repo.