Professional workComputer vision · Wellness

Body Tracker

In-browser face, hand, and pose tracking that turns a webcam into live posture and attention scores.

01Why I built it

Measuring how someone sits, moves, or focuses usually meant wearables, dedicated sensors, or uploading video to someone else's server.

Measuring posture or focus usually means wearables, dedicated sensors, or uploading someone's camera feed to a server. A webcam and a browser should be enough.

Questions I had to answer
  • Q1

    Can face, hand and pose models run fast enough inside a browser?

  • Q2

    How do we keep video private by design, not by promise?

  • Q3

    How can other products reuse the same pipeline?

02Solution

MediaPipe face, hand, and pose models run inside the browser, scoring the camera feed live and exposing the same pipeline as an SDK.

03Key decisions
  1. MediaPipe on the device

    Face, hand and pose models run in the browser, so the camera feed never leaves the device.

  2. Scores, not raw landmarks

    Tracking data becomes posture, attention and fatigue scores people can understand — plus session PDF reports.

  3. Ship it as an SDK

    A JavaScript SDK and React hooks expose the same pipeline to other products.

04Product flow
  1. 01Webcam
  2. 02MediaPipe in browser
  3. 03Posture · attention · fatigue
  4. 04Dashboard · PDF · SDK
05Key features
Live posture, attention, and fatigue scoring
Runs entirely in the browser — video never leaves the device
Hand gesture recognition and multi-person tracking
Session PDF reports
JavaScript SDK and React hooks for other products
06Tech stack
Next.jsTypeScriptGoogle MediaPipeComputer Vision
07Looking back
The hard part

Turning noisy, frame-by-frame landmarks into steady, meaningful scores in real time, on ordinary hardware.

What I learned

Privacy can be an architecture decision. If the video never leaves the device, there's nothing on a server to protect.