Sensei reads boxing the way a good coach reads boxing — at the joints. Hip rotation on the cross, guard hand on the chamber, where the lead foot points, what the head does after the shot lands. Amateur-rules-aware, both stances supported.
Sensei doesn't claim to count your jabs accurately — the model is calibrating, and we're honest about it. What it does, well, is read the shape of the motion: the chain from foot to fist, where the timing falls apart, where the guard drops.
Cross, lead hook, rear uppercut. The hip turns first or the punch is arm-only — the kinematics show which.
The single most common amateur leak: the opposite hand drops when the punching hand fires. Frame-anchored.
Where the lead foot points predicts where the rotation can go. A leak here caps every other lever.
What you do after the last punch matters as much as the punches. Step out? Reset? Or sit and admire?
Pose extraction uses MediaPipe BlazePose Heavy running locally on Sensei's Cloud Run containers. Your video never goes through DeepMotion, Move.ai, or any other external motion-capture API.
20–30 seconds of shadow, bag, or mitts. Five minutes. No card.