PhysIQal Performance

Analyse exercise technique through video, tracking reps and identifying biomechanical faults that are difficult to spot manually.

A Next.js frontend sends workout video to a FastAPI backend, where pose estimation extracts body landmarks that are analysed to score reps and technique.

Built the video pipeline, pose-estimation system and exercise-specific analysis, then iterated heavily on inference speed and deployment performance.

Successfully detects reps and produces technique scores alongside specific feedback on issues such as shoulder movement, wrist position and elbow drift.

Cloud inference was significantly slower than local processing, with video decoding and pose estimation becoming major bottlenecks.

Working prototype with video-based exercise analysis, rep tracking and biomechanical feedback.

Improve inference speed and movement analysis so feedback becomes more precise while remaining fast enough for a smooth user experience.