---
title: Nlp3
emoji: 📉
colorFrom: pink
colorTo: yellow
sdk: gradio
sdk_version: 6.11.0
app_file: app.py
pinned: false
license: mit
short_description: medical rag system (nlp assignment 3)
---

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference


# MedRAG — Hybrid RAG Question-Answering System

**NLP with Deep Learning — Mini-Project 1**

A production-grade Retrieval-Augmented Generation (RAG) system over PubMed medical abstracts.

## Architecture

| Component | Choice |
|-----------|--------|
| Corpus | MedRAG/pubmed — 200 abstracts → 500+ chunks |
| Vector DB | Pinecone (Serverless, Hybrid Index) |
| Dense Embeddings | `all-MiniLM-L6-v2` |
| Sparse Retrieval | BM25 (fitted on corpus) |
| Re-ranking | `cross-encoder/ms-marco-MiniLM-L-6-v2` |
| LLM Generation | OpenRouter (`openrouter/auto`) |
| Evaluation | LLM-as-a-Judge (Faithfulness + Relevancy) |
| UI | Gradio on HF Spaces |

## Required Secrets

Set these in **Settings → Variables and Secrets**:

- `PINECONE_API_KEY` — your Pinecone API key
- `OPENROUTER_API_KEY` — your OpenRouter API key

## Files to Upload

- `app.py` — main application
- `requirements.txt` — dependencies
- `bm25_params.json` — pre-fitted BM25 parameters (generated in Colab)