Your patients' home program, executed and instrumented
You prescribe. PoseDuel makes the program happen between visits — as a game a child wants to play — and returns per-rep movement-quality data you can use at the next appointment. It coaches form with AI inside the bounds you set; it never prescribes, diagnoses, or treats.
Request a pilot ↓The problem — from the literature, not our data
Home exercise programmes account for 50–80% of the total therapy a child receives; sessions with the therapist present make up the remainder. The home program is where the dose lives.
Johnson RW et al. BMJ Open. 2018;8(1):e018316.In a double-blind RCT, a home programme produced significant functional gains at roughly 17.5 sessions a month, averaging 16.5 minutes each — a dose most families never reach unsupported.
Novak I, Cusick A, Lannin N. Pediatrics. 2009;124(4):e606–14.And the reasons it doesn't happen are well characterised: the three main correlates of non-compliance with prescribed exercise are perceived barriers, lack of positive feedback, and helplessness1; in children specifically, time, setting and negative exercise experiences are the recurring barriers, while positive experiences and social support are the facilitators2. PoseDuel is a direct attempt to engineer the facilitators. 1 Sluijs EM, Kok GJ, van der Zee J. Phys Ther. 1993;73(11):771–82. · 2 Holt CJ et al. J Orthop Sports Phys Ther. 2020;50(9):503–15.
Measurement you can interrogate
Every rep is detected and scored for movement quality (tempo, depth, control) against per-exercise criteria. On our recorded evaluation corpus, 97% of reps are accurately detected and scored.
Scoring that nudges, in real time. The engine calibrates to each child: it takes their recent real reps, fits per-criterion bounds that are challenging but reachable, and gives real-time positive feedback when the child stretches to the end of those bounds — a gentle, continuous nudge toward better movement quality, tuned to where that child actually is this week.
What this evidence is The 97% comes from a recorded corpus of roughly 170 sessions of essentially one child (n≈1) — the founder's son, the product's first patient. It measures rep detection; agreement of the movement-quality grades with clinician ratings is exactly what the pilot phase is for. It demonstrates the method, not population-level validity.
How we keep these numbers honest
- Replay-gated evaluation: every change to detection or scoring is replayed against the full recorded corpus before it ships — recall and phantom rates must not regress, per exercise. The metrics above are reproducible runs, not one-off benchmarks.
- Dual-camera capture: a laptop webcam plus a phone from the side. Two views triangulate depth and joint angles a single camera physically cannot see — sagittal-plane quality (squat depth, knee travel) is measured, not guessed.
- Hard to game: a child will optimize whatever scores — so only quality-gated reps power the game (a shallow squat fires nothing), the criteria are ones you set per patient, and the exemplary clips let you audit what the scores were actually given for.
The current exercise catalogue
Eleven movements today, each with per-rep quality criteria (depth, control, tempo, alignment) and difficulty tiers:
- Squat
- Front lunge
- Lateral lunge
- Hip abduction
- Standing crunches
- Plank arm raise
- Plank leg raise
- Plank bird dog
- Standing jog
- High knees
- Jumping jacks
If a program needs a movement we don't cover yet, that conversation is precisely what the pilots are for.
How the workflow runs
- The parent invites you — granting access on their terms (revocable, logged).
- You set the program — which of the supported exercises, targets and difficulty tiers match your prescription. Setup is a few minutes per patient. (Without a clinician, families get sensible defaults — the product is built around your loop, not instead of it.)
- You see quality trends per exercise and criterion over weeks, plus a small set of exemplary clips selected to show how the movement actually looks — not raw session video. It's built to fit the minutes before an appointment, not to add a portal chore.
PoseDuel never issues clinical alerts and the dashboard is pull-only — you are not expected to monitor anything between visits. Families are told plainly: pain or regression means contact your physio.
Data handling, briefly
- All data lives in Europe (Google Cloud, europe-west1) — no ads, no data sales, ever.
- Camera video is uploaded for scoring under the family's privacy settings; parents can have everything deleted, anytime.
- Model evaluation and retraining run on skeleton landmark data, not video.
- Clinician access is parent-granted, scoped to that child, revocable, and logged.
Pilot with us
We're partnering with a small number of pediatric practices for the first clinical pilots. A pilot is deliberately small — a handful of your patients, a short setup call, and we shape the rest together. Leave a practice email and we'll get in touch — or write to hello@poseduel.com.
We reply personally — no mailing list, no spam.