S
Situation
A fitness unicorn's coaching feature required round-trip inference, causing 900ms feedback delays that broke workout flow.
— Project N°06 / Mobile
Kepler Mobile
Cross-platform app with on-device ML coaching.
N°06 · MOBILE
- Client
- Fitness Unicorn
- Year
- 2023
- Duration
- 20 weeks
- Team
- 6 engineers
— STAR breakdown
T
Task
Ship an offline-capable, cross-platform coaching experience with sub-80ms feedback.
A
Action
- 01Ported the coaching model to CoreML + TFLite via a shared Rust inference core.
- 02Rebuilt the client in Flutter with a shared Rust FFI layer for pose estimation.
- 03Rolled out staged releases with on-device telemetry and remote model updates.
R
Result
- 58ms median feedback latency, fully offline.
- Retention up 27% at 30 days.
- App Store featured in 14 countries.
— Stack
FlutterCoreMLRust