PulseMind — demo

Clinical decision support for ICU mechanical ventilation: stream ventilator telemetry, score it for respiratory deterioration within 6 hours, and show a clinician why. Read-only and clinician-in-the-loop. It never controls a ventilator and never recommends treatment.

This repository is the live demo — the only place the trained model reaches a screen.

front-end/   Vite :5173    React dashboard. Reads /api, never derives a band from a score.
back-end/    Node :3500    Express + Mongoose -> MongoDB Atlas. The history of record.
             FastAPI :8000 pythonService/. The only process that touches the model.
contract/                  The shared data contract, as TypeScript types.
checks/                    Executable end-to-end verification.

The model itself lives in a separate repository and is imported as a package; this one owns the serving path and the screen.

Running it

Three processes, in this order. Full commands in back-end/README.md.

..\.venv\Scripts\python.exe -m uvicorn app:app --app-dir back-end/pythonService
node server.js          # from back-end/, needs .env
pnpm dev                # from front-end/

Then POST /api/ward/seed — the board is empty until you do.

Local only. The model service needs a GPU and a 7B language model, so there is no deployed backend. A static deploy of the frontend renders the shell and then shows its error state on every fetch; that is intended.

Two things worth knowing before reading the code

Mongo is not a cache. Each stored reading carries the risk band that the hysteresis machine published at the time. That band is not a function of the reading's own score, so it cannot be recomputed later — losing the collection changes what the ward says, not how fast it says it.

One bed publishes no score at all. None of its eleven parameters arrive, so every input is a population default and the reading falls below the data-sufficiency floor. That is a designed demonstration, not a bug: the system declines to answer rather than answering from statistics about other people.

Verification

checks/ holds the end-to-end tests. There is no CI — verification means running them and reading the output. See checks/README.md.

Data

Everything on screen is manufactured. The model was trained on MIMIC-IV under a PhysioNet DUA; no credentialed or patient-derived data appears in this repository and none may be added to it. .gitignore blocks CSVs outright for that reason.

Status

No authentication, no RBAC, no audit log, and no HL7 feed — all named blockers before any shadow or pilot deployment, all deliberately visible rather than stubbed.