On-device attention signals

OverSight processes eye and face landmarks locally to estimate whether decision-critical information was visually inspected.

Video never leaves this device.

• Runs in this browser tab (MediaPipe, WebAssembly). Frames are read in memory and reduced to a few numbers. Nothing is uploaded or stored.
• No facial recognition, no identity, no age, gender or emotion inference, no fatigue diagnosis.
• Derived signals only: estimated gaze point, dwell per request region, head orientation, face present.
• Prefer not to? OverSight works without a camera using interaction timing, with manual acknowledgement for critical consequences.