---
title: CureMind
emoji: 🩺
colorFrom: blue
colorTo: indigo
sdk: docker
pinned: false
---

# CureMind — AI Medical Information Assistant

An end-to-end **Retrieval-Augmented Generation (RAG)** medical chatbot built as part of a German MSc AI portfolio. It combines curated open biomedical datasets with user-uploaded documents to answer clinical questions with source citations.

---

## Architecture

```
User Question
     │
     ▼
Streamlit UI  ──►  FastAPI Backend
                        │
              ┌─────────┴──────────┐
              ▼                    ▼
    Google GenAI             Pinecone Vector DB
    Embeddings (768d)   ◄──  (medicalindex)
              │
              ▼
     Top-5 Relevant Chunks
              │
              ▼
     ┌────────────────────┐
     │  LLM Router        │
     │  ├─ Qwen-3.8-27B   │  (fast queries)
     │  ├─ GPT-OSS-120B   │  (reasoning / "why/how")
     │  └─ Groq-Compound  │  (fallback)
     └────────────────────┘
              │  (via Groq API)
              ▼
     Answer + Source Citations
```

---

## Tech Stack

| Layer | Technology |
|---|---|
| Frontend | Streamlit |
| Backend | FastAPI + Uvicorn |
| Vector Store | Pinecone (768-dim, dotproduct) |
| Embeddings | Google Generative AI (`embedding-001`) |
| LLM Inference | Groq API (Qwen-3.8-27B · GPT-OSS-120B · Groq-Compound) |
| RAG Framework | LangChain LCEL |
| Datasets | HuggingFace Hub |
| Config | Pydantic Settings |
| Deployment | HuggingFace Spaces (Docker) |

---

## Knowledge Base — Open Datasets

| Dataset | Source | Size | License |
|---|---|---|---|
| PubMedQA | `qiaojin/PubMedQA` | 1 k (labeled) | MIT |
| Mental Health Counseling | `Amod/mental_health_counseling_conversations` | 3.5 k | RAIL-D |
| Medical Meadow MediQA | `medalpaca/medical_meadow_mediqa` | 2.2 k | Academic |
| MedQA-USMLE | `GBaker/MedQA-USMLE-4-options-hf` | 12.7 k | CC-BY-SA-4.0 |

Users can also upload their own PDF documents via the sidebar to extend the knowledge base.

---

## API Endpoints

| Method | Path | Description |
|---|---|---|
| `GET` | `/health` | Service health check |
| `POST` | `/ask/` | Submit a question, returns answer + sources |
| `POST` | `/upload_pdfs/` | Upload PDF documents for indexing |
| `GET` | `/hf_datasets/` | List available HuggingFace datasets |
| `POST` | `/hf_datasets/load/` | Load a dataset into the vector store |

Interactive docs available at `/docs` when running.

---

## Local Development

### 1. Clone and install

```bash
git clone https://github.com/YOUR_USERNAME/curemind.git
cd curemind
```

### 2. Configure environment

```bash
cp server/.env.example server/.env
# Fill in your API keys in server/.env
```

Required keys:
- `GOOGLE_API_KEY` — [console.cloud.google.com](https://console.cloud.google.com)
- `GROQ_API_KEY` — [console.groq.com](https://console.groq.com)
- `PINECONE_API_KEY` — [app.pinecone.io](https://app.pinecone.io)
- `HUGGINGFACE_HUB_TOKEN` — [huggingface.co/settings/tokens](https://huggingface.co/settings/tokens)

### 3. Run backend

```bash
cd server
pip install -r requirements.txt
uvicorn main:app --reload
```

### 4. Run frontend

```bash
cd client
pip install -r requirements.txt
streamlit run app.py
```

---

## HuggingFace Spaces Deployment

### Environment Variables (set in Space Settings → Variables and Secrets)

| Key | Type |
|---|---|
| `GOOGLE_API_KEY` | Secret |
| `GROQ_API_KEY` | Secret |
| `PINECONE_API_KEY` | Secret |
| `HUGGINGFACE_HUB_TOKEN` | Secret |
| `PINECONE_INDEX_NAME` | Variable — `medicalindex` |
| `PINECONE_REGION` | Variable — `us-east-1` |

The Docker container starts both FastAPI (port 8000, internal) and Streamlit (port 7860, public) automatically.

---

## Project Structure

```
curemind/
├── Dockerfile                  # HF Spaces Docker build
├── docker-entrypoint.sh        # Starts both services
├── render.yaml                 # Render.com alternative deployment
│
├── server/                     # FastAPI backend
│   ├── core/
│   │   ├── settings.py         # Pydantic BaseSettings
│   │   └── schemas.py          # Request / response models
│   ├── modules/
│   │   ├── llm.py              # LangChain LCEL chain + model router
│   │   ├── load_vectorstore.py # PDF ingestion pipeline
│   │   └── hf_dataset_loader.py# HuggingFace dataset ingestion
│   ├── routes/
│   │   ├── ask_question.py     # POST /ask/
│   │   ├── upload_pdfs.py      # POST /upload_pdfs/
│   │   └── load_hf_datasets.py # GET|POST /hf_datasets/
│   └── main.py
│
└── client/                     # Streamlit frontend
    ├── app.py                  # Entry point
    ├── components/
    │   ├── chatUI.py
    │   ├── upload.py
    │   ├── hf_loader.py
    │   └── history_download.py
    └── utils/
        └── api.py
```