How on-device AI works

Vision that runs in your tab, not in our cloud.

ScreenMedic's models live inside your browser. Frames are analyzed in real time and discarded immediately — we don't have an upload endpoint for video.

1 · Camera frame

Your browser captures a webcam frame into a sandboxed <video> element. We never copy it to a server — there is no upload endpoint.

2 · On-device inference

Pose landmarks and gaze direction are extracted on your own machine — in the browser tab for Demo Lab, inside the desktop app for the full product.

3 · Metric extraction

From landmarks we derive simple numbers: head angle, shoulder line, distance, blink rate. The frame is discarded immediately.

4 · Only metrics persist

We store posture and eye-health scores — never images, never video, never anything that could identify you.

The stack

Open, audited, measurable.

MediaPipe Pose
MediaPipe Face Mesh
TensorFlow.js
WebAssembly + WebGL