| --- |
| title: MediQuery Healthcare AI |
| emoji: 🏥 |
| colorFrom: blue |
| colorTo: indigo |
| sdk: docker |
| pinned: false |
| app_port: 8501 |
| --- |
| |
| # 🏥 Explainable Healthcare QA Chatbot |
|
|
| An intelligent medical question-answering system combining LLM + RAG + XAI. |
|
|
| ## Quick Start |
|
|
| ```bash |
| # Setup |
| python -m venv venv |
| source venv/bin/activate |
| pip install -r requirements.txt |
| |
| # Download data |
| python scripts/download_data.py |
| |
| # Build knowledge base |
| python scripts/build_knowledge_base.py |
| |
| # Run API |
| python api/main.py |
| |
| # Run Frontend |
| streamlit run frontend/streamlit_app.py |
| ``` |
|
|
| ## Architecture |
|
|
| - **Retrieval**: Hybrid dense (MedCPT) + sparse (BM25) search |
| - **Generation**: Fine-tuned BioMistral-7B with QLoRA |
| - **Explainability**: Confidence scoring, source attribution, rationale generation |
|
|
| ## Project Structure |
|
|
| ``` |
| healthcare_qa_chatbot/ |
| ├── src/ # Source code |
| ├── api/ # FastAPI backend |
| ├── frontend/ # Streamlit UI |
| ├── data/ # Datasets and knowledge base |
| ├── models/ # Trained models |
| └── tests/ # Test suite |
| ``` |
|
|
| ## Components |
|
|
| | Component | Technology | |
| |-----------|------------| |
| | LLM | BioMistral-7B | |
| | Embeddings | MedCPT / all-MiniLM | |
| | Vector Store | ChromaDB | |
| | API | FastAPI | |
| | Frontend | Streamlit | |
|
|
| ## Success Metrics |
|
|
| - Retrieval: Recall@10 > 85% |
| - Medical Accuracy: > 90% |
| - Response Time: < 10s |
| - Safety: < 5% hallucination rate |
|
|
| ## License |
|
|
| MIT License - For educational purposes only. |
|
|
| ## Disclaimer |
|
|
| ⚠️ This is an educational project. The information provided by this system is NOT a substitute for professional medical advice, diagnosis, or treatment. Always consult a qualified healthcare provider. |
|
|
