Kinetra
Track movement.
Understand your recovery.
Kinetra turns any phone camera into a pose-tracking lab — counting reps, scoring smoothness and symmetry, and fusing it with how you actually feel, without a single wearable.
Where generic fitness apps stop.
Rep-counters tell you what you did. Kinetra tells you how well you did it, and whether that's changing over time.
Camera-only pose tracking
MediaPipe-based pose estimation runs on a normal video recording — no wearable sensors, no special hardware, no gym equipment.
Per-limb calibration
A one-time calibration flow learns your own observed range of motion per arm, so thresholds are relative to you, not a generic default.
Honest, non-diagnostic coaching
An optional AI coaching layer reads your scores and trends — never your video — and is explicit that it isn't a medical diagnosis.
One camera, six movement tests.
Every mode uses the same pose pipeline, tuned to the joints and rep pattern that matter for that task.
Chair-rise strength & control
Tracks hip and knee extension across each rise, flagging asymmetric weight-bearing and rushed, uncontrolled descents.
- Rep count & tempo per rise
- Left/right knee-extension symmetry
- Descent-control smoothness score
From a single recording to a movement history.
Kinetra App
The Expo / React Native client: guided recording, calibration, a Daily Check-In log, Trend and Insight screens, and a Team Screening mode for coaches running batches of athletes.
Analysis Engine
A Flask + MediaPipe service that turns a video into joint angles, rep counts, and per-task scores — deployed on Render, with an optional server-side AI coaching endpoint that never sees raw video.
Calibrated to your body, not a generic default.
Kinetra Calibration
Most rep-counters assume everyone's "full range of motion" looks the same. Kinetra records your own observed angle range per arm during a short calibration pass, then derives rep-detection thresholds from that — so a smaller natural range isn't scored as a failed rep.
See it in the beta-
01Record a short baselineA few reps of the task at a comfortable range.
-
02Calibrate per armLeft and right thresholds are stored independently.
-
03Analyze against your baselineEvery future session is scored relative to your own range.
-
04Recalibrate any timeRange of motion changes during recovery — your baseline can too.
Per-side persistence
Left and right calibration values are stored and applied independently, so one side's data can never silently override the other's.
Six task profiles
Sit-to-stand, reach, arm raise, gait, balance, and timed-up-and-go each get their own joint focus and thresholds.
Recalibrate anytime
A visible reset in Settings, since a moving target — recovery — needs a baseline that can move too.
Fusing how you feel with what the camera measures.
A 30-second, no-camera check-in feeds a rule-based engine that looks for real patterns — not for reasons to send a notification.
Daily Check-In
A local, no-network log: an overall feeling rating, multi-select soreness by body region, and an optional note. Takes under 30 seconds, and works even on days with no recording.
Insight Engine
Cross-references check-ins against recorded scores: a streak counter, a soreness-to-score correlation per task, and a left/right asymmetry detector — each gated behind a minimum sample size, so it only speaks when there's evidence.
Evidence over engagement
A correlation only surfaces with at least two comparable sore and clear sessions within a 36-hour window, and a real gap between them. No pattern, no insight — on purpose.
"Reach scores are lower on days you log shoulder soreness — worth mentioning at your next check-in."
"Right-side scores are outperforming left by a wide margin across your last sessions."
Currently a one-person build, in real testing.
Kinetra is an independent project — no company, no funding round, just a working pipeline from a phone camera to a movement score, running on real hardware. The source is open for anyone who wants to read it, fork it, or point out what's wrong with it.