"AI that reads sign language" sounds like a research lab and a GPU cluster. But a genuinely useful starting version runs entirely in your browser, with no model upload and no cloud — the camera feed never leaves your machine. Here's how I built a webcam sign reader from scratch.
This is Day 7 of SolveFromZero, where I solve a real, useful problem each day.
The browser can track a hand
You don't need a server or a camera SDK. Google's MediaPipe ships a tiny hand-tracking model that runs on WebAssembly right in the tab. Hand it a video frame, get back the hand's skeleton — all on-device.
import { HandLandmarker, FilesetResolver } from
