Screens you control with your body, from an ordinary webcam.
I came to software from corrective exercise therapy and front-end web work, and I build with a team of AI coding agents whose work I check. What I make is camera input for the web: body and hand tracking that runs on the device, smoothed into steady pointer input that games and ordinary websites can use. Below: PosePointer, a Chrome extension that turns hand gestures into real mouse input; a booth piece you can try with your own brand name; and an anatomy mirror that shows the muscles moving under your hands.
What I'm building: pose and hand tracking as an everyday input for event booths, kiosks and laptops, so existing screens can be used without touching them. I'm open to booth and kiosk pilots, to creative technologist roles, and grateful for advice from people who have done this work.
Email
Showcases · try them in your browser
Anatomy mirror
Muscles and bones follow your hands on your own webcam.
40 seconds, no sound. Every cursor comes from a real PosePointer run on my recorded or live hand, and each shot says which. Opening footage and demo-page credits are on the end card.
Problem
Kiosks, trade-show product viewers and presentations want touch-free control. Most hand-tracking demos only work inside their own custom app, and synthetic script events are ignored by many real sites.
Solution
A Chrome extension. Hand tracking runs locally in a side panel. Pinch to press and drag, spread two hands to zoom, hold an open palm to pause. The gestures reach the open Chrome tab as trusted mouse input, so an existing 3D product viewer or slide deck responds with no change to the site. It has no code that reads page content or uploads video.
Result
In the first version's replay checks, 11 of 11 pass in Chromium and all 199 page events arrived as trusted input. A pinch drag orbits a model-viewer product and a two-hand spread zooms it. Two pinch clicks advance a reveal.js deck by two slides. A lost hand releases the drag on the first missing frame.
Now
Version 0.3 adds double and right click, one cursor per hand, two-finger touch, and the full-screen anatomy mirror below. It is in private testing and not yet in the Chrome Web Store.
camera → 21 hand landmarks → pinch hysteresis 0.045 / 0.065 → One Euro + 20 ms prediction → Input.dispatchMouseEvent → isTrusted: true
My own movement, from an earlier webcam recording fed into the unchanged demo as its camera. Only the tracked skeleton is shown, never the video.
A brand's logo, or just its name, becomes a field of particles. With the camera on, the body tracked by the webcam pushes them aside, and raising both hands bursts the logo. A cursor or finger works too. When nobody is there, a ghost hand keeps it moving so the screen never sits still. It runs in the browser; nothing is uploaded.
Opens the demo in a new tab. To use a real logo file, drop it on the demo page with "Try your logo"; it stays in your browser.
For a booth it needs a laptop with Chrome or Edge, a webcam and a screen: no sensor or special hardware. PosePointer can run beside it, so visitors can also browse your product viewer or slides without touching the screen. Booth pilots are quoted per event.
This one comes from my years in corrective exercise therapy. The webcam tracks each hand, and a 3D model of the hand and forearm muscles and bones follows it, laid over the live picture as a mirror. A slider fades the camera image so you see more anatomy and less of yourself. It is part of PosePointer 0.3 and runs on the device; nothing is uploaded.
A second, simpler view estimates how far each finger joint bends, in degrees, from the same tracking.
PosePointer (first called Touchless Bridge): the camera signal cleaned up into standard pointer input, so ordinary websites can use it, plus the anatomy mirror.
Also
Addition-chain search for elliptic-curve scalar inversion (inverting modulo the curve's group order). One result: a 284-step chain for P-256 (251 squarings and 33 multiplications), confirmed by two separate checker programs.