build · 2026 · On-device ML product
AI Boxing Trainer
On-device pose coaching with combo curriculum
Flutter · ML Kit · coach logic · Fitness / sports experiment
33 landmarks
Pose
On-device
Mode
Combo FSM
Drills
Flutter ML
Stack
Challenge
Most fitness apps play videos. Real coaching needs live form feedback on-device — without shipping every frame to the cloud.
Approach
- 01Streamed camera frames into Google ML Kit 33-landmark pose detection.
- 02Built biomechanics math + coach FSM for stance and punch states.
- 03Added overlay painter and optional voice coach cues.
- 04Structured combo curriculum (jab → cross → slip → hook) with orthodox/southpaw toggle.
Outcomes
- ▸On-device ML product instincts (latency, permissions, overlay UX)
- ▸Clear software architecture across vision → coach → session modules
- ▸Differentiated Flutter work beyond CRUD apps
AI Boxing Trainer is where mobile engineering meets sports feedback loops.
Keeping inference on-device matters for privacy and speed — the architecture reflects that.
It’s a strong conversation piece for teams exploring camera ML in Flutter.
Stack
FlutterML Kit PoseCameraFSMAudio cues
Exploring on-device vision products? I can adapt this pose → coach pipeline to your sport or form.