A biophysics web app for medical schools: it reads an ECG from a photo, solves problems step by step and explains phenomena at the level the student asks for — from first year to exam prep.
What is inside
- Six modules — ECG, problems, explanations, a Nernst calculator, stats and history; they switch by tab, each keeping its own state.
- Anonymous cookie session — the tool is meant to work without signing up, while a JWT attaches history to an account and enables sharing.
- Rate limit — 30 AI requests per hour per session, an in-memory window on a single uvicorn worker: LLM calls cost money, so the cap sits on the endpoints rather than only in the UI.
- Safety guard — questions outside medicine and biophysics are rejected on the way in.
- PDF export through fpdf2 with a Unicode font for Cyrillic, plus a public result link by token.
- Formulas render with KaTeX — both in LLM answers and in the calculator.
Stack
FastAPI, async SQLAlchemy 2.0, Alembic and PostgreSQL on the backend; React +
Vite + TypeScript and Tailwind on the frontend. The model is called through an
OpenAI-compatible proxy, and the key lives only in the server-side .env.
Demo
biophyslab.ru — no sign-up needed; history is tied to the cookie session.