I’ve got a series of full-length tennis matches archived in MOV format and I want to plug an AI layer on top so I can quickly read how each player really performed. The core of the job is building (or wiring together) a solution that automatically ingests those MOV files, detects every point, and returns clear statistics. The focus is strictly on player performance, not general broadcast highlights or biomechanics. Metrics that must come back from each processed match: • Serve accuracy for each player, split by first and second serve • Rally length, point by point and averaged per set • Shot types with their individual success rate (forehand, backhand, volley, smash, etc.) A summary table or JSON output per match is fine as long as the numbers are reliable. I’m open to any mix of computer-vision frameworks—OpenCV, TensorFlow, PyTorch—or an existing tennis-specific API if that speeds things up. Just make sure the workflow starts with my raw MOV files and ends with the metrics above, delivered in a repeatable way (script, Docker image, or small web app). Please outline: 1. Your proposed tech stack and how you’ll detect serves, rallies, and shot types. 2. Expected accuracy after your first training run. 3. How I’ll run the pipeline locally once you hand it over. If you already have a pretrained model or previous tennis work to show, that will move the conversation along quickly. The systems should run in a Shopify