All software

ParkinPal

A phone web app that measures hand tremor and voice changes for people with Parkinson's, and turns a week of tests into a summary for a doctor visit.

When
February 2026
Type
Team, Cursor AI Hackathon Hamburg, 2nd place
Stack
JavaScript, Web Audio, DeviceMotion, Vercel
Links
Live app, demo video, code
A person reads the test phrase aloud into a phone; beside them, the app's voice test screen shows the phrase, the words it heard and a countdown. Watch the demo on YouTube
From the hackathon demo. The voice test asks for one fixed sentence, transcribes it live and counts the words it recognised.

Tremor test

Thirty seconds of accelerometer data from the phone's DeviceMotion API, high-pass filtered, then checked for energy in the 4 to 6 Hz band typical of Parkinsonian tremor. The result is a score from 0 to 10 and a severity band.

Voice test

A seven-second recording of a fixed sentence through the Web Audio API. The app measures speaking time, pauses of 0.3 s or longer and volume variance, and scores them from 0 to 10. If speech recognition hears fewer than 7 of the 9 words, it asks for a retake rather than scoring a bad recording.

Raw audio and motion never leave the phone. Only aggregated scores go to the model that drafts the report.

Data and the report

Every entry lives in the browser's local storage. For the doctor report, the app sends seven days of daily scores, averages, ranges, standard deviation, trend and the correlation between tremor and voice to a language model through a serverless proxy, and shows a template summary if the model is unavailable.

Limits

DeviceMotion needs HTTPS and a tap to grant permission on iOS, and speech recognition varies by browser. The scores are hackathon heuristics built in a weekend. They track change for one person over time; they are not a clinical measure.