We built an AI tennis analysis agent. The frontend sends a user_id and a short video URL, then our backend downloads the video, uses FFmpeg to extract key frames, and sends those frames to a vision model for tennis-specific analysis.
The system identifies things like stroke type, movement phase, footwork, strengths, issues, and coaching tips. Then it converts that structured analysis into a natural language text response that the frontend can directly use for voice playback.
We also connect the result with user context and memory, so the feedback becomes more personalized over time instead of treating every video as a one-off clip.
Log in or sign up for Devpost to join the conversation.