Featured project
BVAI
Markerless pose estimation and biomechanical scoring for baseball swings, on video shot with a phone.
Live with usersLive on the web at Bobby Valentine’s Sports Academy. The iOS app is on TestFlight, ready for and awaiting App Store release.
My role: Solo · AI Product Lead
The problem
Turn phone-recorded baseball swings into feedback coaches can act on, and answer athlete questions from the academy’s own coaching material.
What I built
- Built the coaching platform solo for Bobby Valentine’s Sports Academy; it is live with users on the web, with iOS on TestFlight awaiting App Store release.
- Connected markerless pose estimation to biomechanical scoring that flags deviations in a swing.
- Added curriculum-based retrieval and multi-provider model routing with cost, latency, and fallback handling.
Engineering decisions
- Use parity tests across model families to catch disagreements between their different keypoint layouts.
- Ground athlete answers in academy material rather than relying only on general model knowledge.
- Share the TypeScript application across web and iOS through Capacitor.
Verification
- Golden-frame and jitter regression tests check pose-pipeline behavior.
- A golden-retrieval evaluation suite checks curriculum retrieval.
- More than 30 CI suites and device-verification gates guard production changes.
Tech stack
- TypeScript
- Express
- Drizzle
- RTMPose
- MediaPipe
- MotionBERT
- Pinecone
- Capacitor
- Stage
- Live with users
- Role
- Solo · AI Product Lead
- Platform
- iOS (Capacitor)
- Year
- 2025–present
