Every attempt is a short video of four progressive shapes. The platform turns it into four numbers — and a percentile that tells a trainee whether they're at trainee, competent, or expert level.
A trainee records an attempt, uploads it, and gets a validated AI report within minutes. The video below is a real Shape 1 attempt on the platform.
For comparison, here is the same Shape 1 task scored in the Expert tier. Same shape, same metrics — different result.
Trainee workflow: record → upload → AI scores within minutes → validate or contest. All attempts are stored against the trainee profile so improvement over time is visible.
Four metrics, each derived directly from the recorded video. Together they form the percentile and tier.
Distance the tip travels per unit active time. Reported in two flavours so trainees see the difference between raw movement and movement with intent.
Share of frames where tip jitter falls below the expert threshold. Decomposed into three bands so trainees can see where stability breaks down.
Fraction of active time the tip lies within the stencil's blue track. A pure accuracy metric — were you where you were supposed to be?
Total Test Time vs Active Intention Time. The gap between them reveals hesitation patterns invisible to the naked eye.
No black box. Four stages, each with a defined input and output.
Video sampled at 30 fps and masked to the endoscope field of view. Vendor overlays are ignored.
A custom computer-vision model trained on ham-specific imagery locates the scope tip in every frame.
The blue stencil is segmented; the distance from the tip to the nearest line is computed per frame.
Per-frame data is aggregated into the four metrics, then z-scored against a reference cohort of experienced endoscopists.
Each metric is z-scored against a reference cohort of experienced endoscopists. The z-score becomes a percentile, and the percentile lands the attempt in one of three tiers.
Tiers are computed per shape, because shape difficulty varies. A trainee can be Expert on Shape 1 and Competent on Shape 4 — and that's the data the platform surfaces.
HAM stratifies trainees from interventional specialists. AI scoring shows comparable performance to human raters. Three peer-reviewed studies — and a featured talk at ESGE Days 2026 — back the method.
Read the research →