A safety-first clinical AI backend that combines medical image analysis, database-backed clinical research, and retrieval-augmented reasoning.
Designed strictly for clinical decision support and research, not diagnosis or treatment.

This system does not provide medical advice, diagnosis, or treatment.
All outputs are informational and must be interpreted by qualified healthcare professionals.

Most healthcare AI systems suffer from one of two issues:

• Black-box predictions with no transparency
• Unsafe diagnostic claims without grounding in real clinical data

This project addresses both by:

• Using domain-trained medical models
• Querying real hospital data (MIMIC-IV)
• Enforcing read-only SQL
• Providing evidence-based reasoning
• Clearly separating analysis from decision-making

• DenseNet-121 pretrained on MIMIC-CXR / CheXpert
• Smart lung cropping
• Test-time augmentation (TTA)
• Probability calibration
• Lay-language explanations
• /api/cxr/predict

• Intent classification:
  • Statistical → SQL aggregation
  • Reasoning → Case-based retrieval
• SQL templates stored in MongoDB (vector search)
• Auto-tuned SQL using LLMs
• Read-only guards + auto-limits
• /api/research

• Vector search over real hospital cases
• Similar-patient retrieval
• Multi-agent reasoning (diagnostician, differential, clinician)
• Evidence-linked outputs (patient IDs)
• JSON-structured results
• /api/diagnose

• Backend: FastAPI
• Vision: TorchXRayVision, PyTorch
• NLP: SentenceTransformers, Gemini, Azure OpenAI
• Databases:
  • PostgreSQL (MIMIC-IV)
  • MongoDB Atlas (vector search)
• Safety: SQL guards, no-write enforcement
• Deployment: Uvicorn, Procfile

clinical-ai-assistant-backend/
├── main.py              
├── requirements.txt     # dependencies only
├── .env.example         # environment variables template
├── Procfile             # deployment entry
├── README.md            # documentation
├── LICENSE             
└── .gitignore           

pip install -r requirements.txt
uvicorn main:app --host 0.0.0.0 --port 8000 --reload

Clinical research queries Hospital data exploration Medical AI prototyping Decision-support system demos AI safety demonstrations in healthcare