← Back to Projects
Hospital Center AI
9 AI specialists triaging emergency patients
LangGraph PostgreSQL WebSocket FastAPI D3.js
Screenshots
Architecture
Patient -> Triage -> [8 Specialists in Parallel] -> Consensus -> Selected Specialist -> Conversational ChatOverview
A patient describes symptoms. The triage agent analyzes urgency and routes to 8 medical specialists evaluating simultaneously. A consensus agent selects the best match, then the patient enters a conversational chat with that specialist -- all backed by persistent PostgreSQL memory and LangGraph checkpointing.
Specialists: General Medicine, Cardiology, Neurology, Pediatrics, Dermatology, Traumatology, Psychiatry, Oncology.
48 source files, ~6300 LOC. Strict mypy, parametrized SQL (zero injection surface), retry with exponential backoff, structured logging, JWT + HMAC cookie auth, HIPAA/GDPR consent flow, non-root Docker containers, 70%+ test coverage enforced.
Tech Stack
| Layer | Technology |
|---|---|
| Orchestration | LangGraph 1.0 (parallel state machine) |
| LLM | Groq API (Llama 4 Scout / OpenAI-compatible) |
| Backend | FastAPI + Uvicorn (async, 4 workers) |
| Database | PostgreSQL 15 (conversations + checkpoints) |
| Real-time | WebSocket (Socket.IO) |
| Auth | JWT + HMAC session cookies |
| Frontend | Jinja2 + D3.js graph visualization |
| Deploy | Docker Compose (multi-stage, non-root) |