How AI video analysis actually reads a BJJ roll

When you upload a sparring video, the engine doesn't just look at pixels. It runs the footage through several specialized layers — pose estimation, position recognition, transition detection, and submission heuristics — then assembles the results into a single, timestamped narrative.

The first layer locates each athlete in the frame and tracks their relative position over time. From there, a position classifier asks: is this guard, half guard, side control, mount, or back?

The transition layer watches for the moments those positions change. A sweep, a pass, an escape — each one is a labeled event with a timestamp.

Finally, a submission detector flags chokes, joint locks, and attempts. The output is a scene-by-scene breakdown that you can click through to jump to the right moment in the video.

The result: instead of rewatching a 6-minute roll three times trying to remember when something happened, you get a coach's notebook handed back to you in under two minutes.