Workout classifier
Passive detection labels the lift in milliseconds, aligns sets to exercises, and keeps your log truthful without distracting you mid-set.
Passive detection • active
System architecture
The intelligence layer
TrueForm turns commodity smartphone sensors into edge inference: sub-50ms coaching loops with privacy-preserving on-device processing and federated-compatible telemetry when you choose to sync.
Inference breakdown
Muscle activation
Temporal attention over pose streams surfaces which tissues drive each rep versus which ones hitchhike — a live map of workload quality.
Mapping • streaming
Form correction
A control-theoretic cueing layer resolves joint-angle targets into actionable on-screen arrows while you move — correcting before the bad reps stack.
Mid-rep cues • locked
Runs where you lift
No cradle, no beacon, no calibrated rig. Inference is quantized for modern phone NPUs/GPUs, thermally guarded for long sessions, and built to degrade gracefully under motion blur or crowded backgrounds.
- On-device by default • optional encrypted cloud history
- Works offline for core cues when airplane mode kicks in
- Designed for commercial gym entropy, not pristine studios