The intelligence
of muscle.
Passive detection keeps context honest. Activation mapping explains load quality. Correction closes the loop while motor patterns still adapt — all runnable on commodity phone sensors.
Latency profile
<50ms
Humans in charge. Models as spotters.
TrueForm cues are probabilistic overlays — not diagnoses. Coaches, clinicians, and athletes stay accountable for load selection, pathology, and long-term progression. We prioritize transparent limitations, calibrated language, and opt-in telemetry.
- ● No automated medical claims.
- ● On-device inference as default posture for biometric video.
- ● Auditable cue phrasing grounded in kinematics, not shame.
Research-grade habits, consumer ergonomics
We fuse classical biomechanics with modern temporal models that respect occlusion, jitter, and bad gym lighting — because sterile lab accuracy that fails on Friday night squats ships zero value.
Dataset hygiene
Curated tagging, disagreement review, rolling validation against fresh athletes.
Responsible iteration
Staging rings, rollback paths, telemetry budgets for corrections.
Unified inference
Detection aligns context. Activation isolates leaks. Correction times cues to peri-rep stiffness changes so feedback lands when athletes can actually use it.
