An AI workout partner that counts your reps and checks your form from any webcam — no wearables, no straps, no extra hardware. It reads your body in real time, knows a clean rep from a half one, and tells you the moment your form slips.

Training alone means counting in your head while a wandering mind drops the tally at rep seven — and no one's watching the half-reps, the dropped elbow, the squat that never hit depth. Wearables and smart-gym rigs promise to help, but they mean strapping on hardware and trusting a number you can't see.
The brief: turn the webcam everyone already has into a rep counter and form coach. Read the body live, count only the reps that complete a full range, and call out form the instant it slips — all on screen, nothing worn.
MediaPipe Pose tracks 33 body landmarks on the live webcam feed every frame — shoulders, elbows, wrists, hips, knees, ankles — giving the system a full skeleton to reason about, no markers or wearables needed.
Each exercise has its own joint-angle state machine. A bicep curl only counts when the elbow travels from extended to fully flexed and back — so half-reps and bounces don't inflate the number. Six exercises, each with its own thresholds.
The same joint angles drive live form checks — a curl that doesn't finish, a squat short of depth, elbows flaring on a press. OpenCV draws the skeleton, rep count, and feedback straight onto the feed, and the whole session saves as a video with overlays baked in.
I train at home with no one watching. Now my laptop counts every clean rep and tells me the second my form slips — it's like having a coach standing right there.
Bicep curls, squats, push-ups, shoulder press, lateral raises, and lunges — each with its own joint-angle state machine and form rules.
Full-range-of-motion thresholds mean partial reps and bounces don't inflate the count — the number you see is the work you actually did.
Everything runs on the webcam you already own. No watch, no strap, no smart-gym rig — open the camera and start lifting.
Rep count, skeleton overlay, and form cues — the full pose-estimation pipeline in one live view.