Padel Shot Classifier
An end-to-end ML system that detects and classifies padel shot types (smash, bandeja, víbora, volley, gancho, bajada, and more) from match video, built from the ground up as a full MLOps pipeline. Covers sourcing and annotating footage from broadcast match replays, a pose-based candidate-detection stage that flags likely shot moments in hours of raw video for review, and a labeled dataset pipeline with full source/camera/split provenance tracking; the model itself is served via a containerised inference API that takes match video as input and returns detected shots with timestamps and confidence scores, with production infrastructure including monitoring for prediction drift and latency and a human-in-the-loop review step feeding corrected labels back into scheduled retraining runs.
Demo link coming soon.