Quick audit of a golf swing analysis pipeline (MediaPipe pose, Python)

I'm building a golf swing analysis tool that works from 2D video (Face On and Down the Line views). It uses MediaPipe Pose in Python to extract keypoints, detect swing phases (address, top, impact, finish) and compute swing metrics. I'm looking for an experienced computer vision engineer to independently review the full codebase and tell me how to make it more robust and accurate. This is a review, NOT a rewrite. I will implement the changes myself. What I will provide: - Access to a private GitHub repository (after NDA) - A README with setup instructions and a short overview of each module - A few sample swing videos to run the pipeline on Scope: - Review the full pipeline and code on your own - Run it on the sample videos - Identify the main weaknesses: keypoint jitter, occlusions (hands/club), phase detection reliability, sensitivity to camera angle and lighting - Tell me which metrics are realistically reliable from 2D video and which are not - Deliver a written report with prioritized recommendations (2–4 pages), pointing to the specific files/functions involved - Optional: a 30-min call to go through the report Ideal profile: - Hands-on experience with MediaPipe / pose estimation - Experience with sports or movement analysis is a big plus Budget: fixed price. If this goes well, there may be follow-up work. Confidentiality: an NDA is required before any code is shared.