TrueForm

Most people train alone and cannot see their own form. A trainer costs $50 to $150 a session, and a mirror does not measure anything, so form mistakes quietly stall progress or lead to injury. We wanted anyone with a laptop webcam to get a trainer's kind of feedback after every set.

Pick one of nine exercises, start the camera, and record a set while a live skeleton overlays your movement. When you stop, the results screen gives a form score, a rep count, the measured joint angles, specific corrections with evidence, and a short coaching note.

A teammate set up the first pose-detection foundation and the reference-video baseline. I built the analysis on top of it: exercise-specific analyzers, including jumping jack, pull-up and sit-up, and a push-up rewrite around body alignment because elbow angle proved unreliable from a webcam. I also wrote the rep counter, the results screen, a fix for the camera staying on after recording, and the AI coaching note.

Pose analysis runs in the browser with MediaPipe, so video never leaves the device. Scoring is deterministic, meaning the same movement always scores the same. The AI only sees the measured numbers, so it can put the feedback in plain words but cannot invent a measurement or change the score. It is live at blunelabs.com/trueform.

Screens captured from a local run using the repository's reference bicep-curl video in place of a webcam.