MediaPipe
MediaPipe is the on-device perception framework that lets Posture Maxing watch you without ever phoning home.
What it is
MediaPipe is an open-source toolkit from Google for building real-time perception pipelines. Pre-trained ML models bundled with a WebAssembly + GPU runtime, the same vision code runs inside an Electron app, a browser tab, or a mobile app.
The Pose Landmarker
Posture Maxing uses the Pose Landmarker, a model that infers 33 body landmarks per frame from a webcam image. We ship the lite variant: about 5 MB on disk, runs at 30+ fps on any GPU made in the last decade.
Only six of the 33 landmarks feed the posture score: ear and shoulder on each side, plus the hips. The rest are drawn on the skeleton overlay for visual feedback but never enter the calculation.
Why local-first
Posture data is biometric. Sending it to a cloud server is a permanent privacy commitment, once it leaves your device, you can't recall it. MediaPipe lets us skip that decision entirely. There's no server to compromise, no API key to manage, no terms of service for your spine.
Browser vs Electron
The same MediaPipe build runs in both contexts. In the desktop app, the .task model file is downloaded once on first launch and cached in the user data folder. In the browser demo on this site, the model is fetched from MediaPipe's CDN on click, identical precision, just no persistence between visits.