A camera "mirror" that reads a shopper's face and points them to the sunglasses that actually suit them — turning "which pair is right for me?" into a guided path to checkout.
Choshmish sells sunglasses online, and online is exactly where buying glasses gets hard — a customer can't try anything on, can't tell which frames flatter their face, and often bounces rather than guess.
AI Mirror closes that gap: a lightweight camera experience that reads face shape and fit, then recommends the frames that work — no measuring, no guessing.
A floating launcher opens a full-screen camera overlay. From there, everything happens in the browser — landmarks in, tags out — before routing the shopper to frames that fit.
MediaPipe FaceMesh hands back 468 facial landmarks in real time. AI Mirror doesn't need all of them — just enough geometry to make two simple calls, then a third to turn them into something shoppable.
tag:oval AND tag:standard — that filter the live catalog, so results are always in-stock, real products.One floating button that drops into any storefront page — no redesign required.
Real-time FaceMesh overlay drawn on a mirrored canvas, so the camera reads naturally.
Derived straight from facial geometry — the height-to-width ratio of the detected face.
Pupil distance relative to face width maps directly to frame fit.
A result card routes the shopper to a tag-filtered search — always in-stock, real products.
All processing happens in the browser — no camera frame is ever uploaded or stored.
The first version worked but was rough. I rewrote it end to end: tightened the face-shape classification beyond the original two-threshold approach, moved all processing on-device so no camera frame ever leaves the browser, and closed the security gaps in how the widget loaded and mounted.
A symmetry score sits in the result as a light UX flourish — not a biometric measurement — for a touch of delight without overstating what the camera can tell you.