JeevaMano/genai-clinical-qa-assistant

🧠 GenAI Clinical Q&A Assistant

A GenAI-powered clinical question-answering system that allows users to ask questions about medical PDFs, patient cases, or discharge notes — and get accurate, natural language answers.

🚀 Features

• —🔍 Semantic search over clinical documents using FAISS & Sentence Transformers
• —💬 Question answering via FLAN-T5 (HuggingFace Inference API)
• —📄 Upload your own medical PDFs for live querying
• —🌐 Frontend built with Streamlit (no React/Node required)
• —🧠 Custom-trained on real medical CSVs converted into PDF
• —⚙️ Modular architecture using FastAPI, Python, and Hugging Face

📁 Project Structure

genai-clinical-qa/
├── backend/
│   ├── app.py                # FastAPI app for backend
│   ├── qa_engine.py          # Embedding + QA logic
│   ├── document_loader.py    # PDF text extraction
│   ├── embedding_store.py    # FAISS vector index
├── frontend/
│   └── app.py                # Streamlit UI
├── data/
│   └── clinical_docs/        # Sample/generated PDFs
├── generate_pdfs_from_csv.py # Create PDFs from medical CSVs
├── requirements.txt
└── .env                      # HuggingFace token (excluded from Git)

🛠 Setup Instructions

# 1. Clone the repository
git clone https://github.com/Jeeva-soul/genai-clinical-qa.git
cd genai-clinical-qa

# 2. Create and activate virtual environment
python -m venv venv
venv\Scripts\activate        # (Windows)
# or
source venv/bin/activate     # (macOS/Linux)

# 3. Install required packages
pip install -r requirements.txt

# 4. Add HuggingFace token to .env
echo HUGGINGFACEHUB_API_TOKEN=your_token_here > .env

# 5. Run the backend
uvicorn backend.app:app --reload

# 6. In a new terminal, run the Streamlit frontend
cd frontend
streamlit run app.py

📬 Example API Request (Backend)

POST /ask
{
  "question": "What medications were prescribed?"
}

💻 Streamlit UI Features

• —Elegant design with custom CSS
• —Real-time question box with loading spinner
• —Upload your own PDFs to instantly augment the knowledge base
• —Beautifully styled answer container

📊 Dataset Used

• —🧠 Brain Tumor Dataset (Kaggle)
• —📄 Converted to PDFs to simulate real-world clinical documents

🧩 Future Enhancements

• —🧬 Clinical entity tagging (ICD-10, SNOMED)
• —☁️ Cloud deployment (Azure, Hugging Face Spaces)
• —🔁 Multiple LLM support (switch between APIs/local models)
• —📈 Session history & PDF content preview

⚠️ Security Note

Do not commit your .env or HuggingFace API keys to GitHub. They are automatically excluded using .gitignore.

👨‍💻 Author

Jeeva Manavalan 🎓 University of Texas at Arlington | Data Scientist | GenAI Enthusiast 🔗 LinkedIn • GitHub

📄 License

MIT License