Add HF Spaces README config

README.md

@@ -1,178 +1,14 @@
-# 🏥 MedRAG — Medical Knowledge RAG System
-
-A full-stack, production-ready Retrieval-Augmented Generation system for medical knowledge, with a ChatGPT-style interface, per-user chat history, JWT auth, and an admin dashboard.
-
-## Tech Stack
-
-| Layer | Technology |
-|---|---|
-| **LLM** | Llama 4 Scout 17B via **Groq** |
-| **Embeddings** | `NeuML/pubmedbert-base-embeddings` |
-| **Reranker** | `cross-encoder/ms-marco-MiniLM-L-6-v2` |
-| **Vector Store** | **Qdrant** (local path storage) |
-| **Backend** | **FastAPI** + SQLAlchemy async |
-| **Auth** | **JWT** (access + refresh tokens) |
-| **Database** | **PostgreSQL** (users, docs, chat sessions, and message history) |
-| **Frontend** | React + Tailwind + Zustand + Framer Motion |
-
----
-
-## Project Structure
-
-```
-medical-rag/
-├── app/
-│   ├── main.py                    # FastAPI app, lifespan, routers
-│   ├── core/
-│   │   ├── config.py              # Pydantic settings
-│   │   └── security.py            # JWT helpers
-│   ├── db/
-│   │   ├── database.py            # Main database (engine, sessions, and core schema)
-│   │   └── chat_db.py             # Chat history models (conversations + messages)
-│   ├── schemas/schemas.py         # All Pydantic models
-│   ├── services/
-│   │   ├── embedding_service.py   # PubMedBERT async encoder
-│   │   ├── reranker_service.py    # Cross-encoder reranker
-│   │   ├── qdrant_service.py      # Qdrant CRUD
-│   │   ├── llm_service.py         # Groq → Llama 4 Scout
-│   │   ├── ingestion_service.py   # PDF/DOCX/TXT → chunk → embed → store
-│   │   └── rag_pipeline.py        # Retrieve → Rerank → Generate → Persist
-│   └── api/routes/
-│       ├── auth.py                # /auth/register, /login, /refresh, /me
-│       ├── conversations.py       # /conversations CRUD
-│       ├── query.py               # /query  (RAG endpoint)
-│       ├── documents.py           # /documents upload/list/delete
-│       └── health_admin.py        # /health, /admin/*
-├── frontend/
-│   └── src/
-│       ├── api/client.js          # Axios + auto-refresh interceptor
-│       ├── store/index.js         # Zustand: auth + chat state
-│       ├── pages/
-│       │   ├── AuthPage.jsx       # Login + Register
-│       │   ├── ChatPage.jsx       # Main ChatGPT-style chat
-│       │   └── AdminPage.jsx      # Admin dashboard
-│       └── components/
-│           ├── layout/Sidebar.jsx # Conversation list, search, grouped by date
-│           └── chat/
-│               ├── MessageBubble.jsx  # Markdown, sources, timing
-│               ├── ChatInput.jsx      # Textarea + PDF upload + button
-│               ├── UploadModal.jsx    # Drag & drop upload dialog
-│               └── WelcomeScreen.jsx  # Suggestion cards
-└── requirements.txt
-```
-
----
-
-## Quickstart
-
-### 1. Clone & configure
-
-```bash
-cp .env.example .env
-# Edit .env — fill in GROQ_API_KEY, SECRET_KEY, and DATABASE_URL (PostgreSQL connection string)
-```
-
-### 2. Backend (local dev)
-
-```bash
-python -m venv .venv
-source .venv/bin/activate        # Windows: .venv\Scripts\activate
-pip install -r requirements.txt
-
-# Start server
-uvicorn app.main:app --reload --port 8000
-```
-
-API docs: http://localhost:8000/docs
-
-### 3. Frontend (local dev)
-
-```bash
-cd frontend
-npm install
-npm run dev
-# → http://localhost:5173
-```
-
 ---
-
-## RAG Pipeline
-
-```
-User question
-    │
-    ▼
-[PubMedBERT Embed]  — medical-domain embedding
-    │
-    ▼
-[Qdrant Search]  — top-20 cosine similarity candidates
-    │
-    ▼
-[Cross-Encoder]  — ms-marco-MiniLM-L-6-v2 reranks to top-5
-    │
-    ▼
-[Llama 4 Scout]  — Groq inference, medical system prompt
-    │
-    ▼
-Answer + Sources — saved to PostgreSQL database
-```
-
----
-
-## API Reference
-
-| Method | Endpoint | Auth | Description |
-|---|---|---|---|
-| POST | `/api/v1/auth/register` | None | Create account |
-| POST | `/api/v1/auth/login` | None | Get JWT tokens |
-| POST | `/api/v1/auth/refresh` | None | Refresh access token |
-| GET | `/api/v1/auth/me` | User | Current user info |
-| GET | `/api/v1/conversations` | User | List user's conversations |
-| POST | `/api/v1/conversations` | User | Create new conversation |
-| GET | `/api/v1/conversations/{id}/messages` | User | Get chat history |
-| PATCH | `/api/v1/conversations/{id}` | User | Rename conversation |
-| DELETE | `/api/v1/conversations/{id}` | User | Delete conversation |
-| POST | `/api/v1/query` | User | RAG query (returns answer + sources) |
-| GET | `/api/v1/documents` | User | List all documents |
-| POST | `/api/v1/documents/upload` | Admin | Upload + ingest document |
-| DELETE | `/api/v1/documents/{id}` | Admin | Delete document + vectors |
-| GET | `/api/v1/admin/stats` | Admin | System statistics |
-| GET | `/api/v1/admin/users` | Admin | All users |
-| PATCH | `/api/v1/admin/users/{id}/role` | Admin | Change user role |
-| PATCH | `/api/v1/admin/users/{id}/toggle` | Admin | Enable/disable user |
-
+title: MedRAG
+emoji: 🩺
+colorFrom: blue
+colorTo: purple
+sdk: docker
+pinned: false
 ---
 
-## Database Schema
-
-All tables (users, documents, conversations, and messages) live in the same PostgreSQL database. Strict user data isolation is enforced by queries filtering on the `user_id` column.
-
-Conversations are auto-titled from the first message and grouped in the UI by Today / Last 7 days / Older.
-
----
-
-## Environment Variables
-
-| Variable | Description |
-|---|---|
-| `SECRET_KEY` | JWT signing secret (use a long random string) |
-| `GROQ_API_KEY` | Your Groq API key |
-| `LLM_MODEL` | `meta-llama/llama-4-scout-17b-16e-instruct` |
-| `QDRANT_PATH` | `data/qdrant` |
-| `QDRANT_COLLECTION_NAME` | `medical-knowledge-rag` |
-| `QDRANT_NAMESPACE` | `medical-docs` |
-| `EMBEDDING_MODEL` | `NeuML/pubmedbert-base-embeddings` |
-| `RERANKER_MODEL` | `cross-encoder/ms-marco-MiniLM-L-6-v2` |
-| `RAG_RETRIEVAL_TOP_K` | Candidates from Qdrant (default: 20) |
-| `RERANKER_TOP_K` | Final chunks after rerank (default: 5) |
-| `RAG_CHUNK_SIZE` | Tokens per chunk (default: 512) |
-| `RAG_CHUNK_OVERLAP` | Overlap tokens (default: 64) |
-
----
+# MedRAG — Medical Knowledge RAG Assistant
 
-## Notes
+An AI-powered medical question answering system using Retrieval-Augmented Generation (RAG).
 
-- **Document upload is admin-only** — regular users query but admins manage the knowledge base
-- **No documents = no answers** — ingest PDFs/DOCX first before querying
-- The first model load (PubMedBERT + cross-encoder) takes ~30–60 seconds; subsequent requests are fast
-- For production: optionally move Qdrant to a managed/cloud deployment (the app is already fully migrated to use PostgreSQL)
+Built with FastAPI, PubMedBERT embeddings, Qdrant vector database, and Groq LLM.