Movement Assessment Prediction System
A webcam-based deep squat assessment system that uses MediaPipe pose estimation and a five-model ML pipeline to check recording quality, identify movement boundaries, classify form with ~92% accuracy, and score movement quality.
01Built over 16 Agile sprints without depth-sensor hardware, keeping inference CPU-friendly.
02The final GRU scoring model achieved 0.062 MAE and 0.67 Pearson correlation against expert ratings.
03Unified the pipeline into one API endpoint so pose extraction runs only once, delivering ~18.5s total processing time with 43 automated tests and CI/CD deployment.
PythonMediaPipeMachine LearningGRUCI/CD