ML Engineer — Player & Ball Tracking from Sports Footage

I am building a sports technology platform that processes game footage to automatically track players, the ball, and referees, and generate performance insights for coaches and players.This is a proprietary product and full details will be shared after an NDA is signed. What I need built: Phase 1 — A stable, reliable player tracking pipeline that ingests raw video, detects and tracks all players and the ball, automatically separates players into two teams, identifies referees separately, and outputs an annotated video plus a CSV of player coordinates per frame. Phase 2 — Optimize the pipeline for speed and cost. Add key moment detection from longer clips. Target: process a 5 minute clip in under 10 minutes on a GPU. Phase 3 — Wrap the pipeline in a FastAPI backend, deploy to a cloud GPU server, and connect it to an existing web frontend so users can upload video and receive results automatically.