I’m building an end-to-end analytics engine that turns raw padel match footage into structured, queryable data. The system has to work equally well with phone-recorded clips and fixed-camera streams, so calibration routines must adapt to differing viewpoints and resolutions. Core functionality I need you to deliver: • Detect and track the four players, the ball, and full court lines, then translate everything into real-world coordinates. • Automatically segment rallies, flag serves and returns, and classify common shot types — forehand, backhand, volley, lob, smash, bandeja and víbora are enough for v1. • Identify winners and errors so each rally ends with a clear outcome. • Produce JSON that contains every event, rally, frame-level positions and overall statistics. Heatmaps don’t need to be visualised yet; the positional data for them must also be supplied in JSON so I can generate either static or interactive charts later. Key statistics that must appear in the output: 1. Player positions and movement traces throughout the match 2. Types of shots taken and whether they resulted in winners, forced errors or unforced errors 3. Rally durations paired with their outcomes Technology preferences are Python with OpenCV, YOLO-based detection, pose estimation for finer tracking, and GPU-accelerated processing on AWS or GCP (or a local CUDA setup if you prefer). A clean, well-documented codebase and brief setup script are part of the hand-off. When you reply, please show: • Examples of previous computer-vision or sports-analytics projects you’ve delivered • A concise outline of the approach you’d take for detection, tracking and event logic • Your estimated timeline from kick-off to first working prototype and to final delivery I’m happy to answer any clarifying questions to help you scope your proposal accurately. Looking forward to seeing how you’d tackle turning padel video into actionable insights.