VisionFiT
- CATEGORYCV posture coaching
- ENGINEPose tracking + rules
- INPUTWebcam, camera-only
- STATUSProof of concept
A real-time AI posture coach — a vision engine that watches your joints, validates your form against tunable biomechanical rules, counts valid reps, and coaches you out loud, rep by rep — then archives the session into deep kinematic charts. Camera-only, runs in the browser. Built by IndiaNIC.
Good form is invisible without a coach.
Training alone means no safety net — bad reps go uncorrected and asymmetries compound. VisionFiT turns a webcam into a second set of expert eyes. It’s a showcase of our computer-vision engineering — explore the rapid-POC program to build one on your own use case.
Camera-only, runs locally
Real-time tracking inside the browser — no wearables, no sensors, nothing leaves the device.
Rules, not vibes
Every rep is judged against explicit, tunable thresholds — angles, ratios, stillness frames — not a vague score.
Eyes-free coaching
An energetic, throttled voice HUD corrects form out loud so you never break posture to glance at a screen.
Bilateral truth
Left-vs-right symmetry plots surface the asymmetries a mirror quietly hides.
A skeleton overlay that judges every rep.
A live neon skeleton tracks your joints, valid reps tick up, and the voice HUD speaks corrections the instant your form drifts — so your eyes never leave the movement.
One vision engine. Four pillars.
Track the body, judge the form, coach by voice, and turn the whole session into reviewable diagnostic data.
- 01 · PILLAR
Kinematics tracking
Monitors joints frame-by-frame, calculating precise angles and mapping skeletal symmetry from a single camera.
- 02 · PILLAR
Form rules engine
Matches live telemetry against physical safety parameters — preventing valgus collapse, excess lean, and spine stress.
- 03 · PILLAR
Voice coaching
Throttled spoken cues turn raw data (“knees beyond toes”) into friendly advice (“bring knees back”), rep by rep.
- 04 · PILLAR
Session diagnostics
Plots bilateral symmetry, movement rhythm, and form-flaw distributions for coaching-grade review.
Frame, train, review.
No wearables and no manual logging — the whole loop from set-up to coaching-grade review happens through one camera.
Frame & calibrate
Pick an exercise, position the camera per its angle badge, hit start; a 5-second countdown lets you step back before the engine locks your stance.
Train with live coaching
Perform reps while the skeleton tracks joints in real time; the voice HUD counts valid reps and speaks corrections the instant form drifts.
Review the breakdown
Every session auto-saves with symmetry charts, tempo plots, phase markers, and a form-flaw distribution for review.
Every movement, validated to spec.
Each exercise carries its own ruleset of tunable thresholds — tightened for elite athletes, loosened for rehab — all on the same engine.
Standard squats
Knee-depth target, anti-lean torso ratio, valgus (cave-in) guard, stance width + foot turnout, and a stillness gate before a rep arms.
Pile squats
Enforces a wide stance, heavy foot turnout, a strictly vertical torso, and knees tracking over toes to load the adductors.
Split lunges
Side-view analysis: lunge depth, front knee kept behind the toe plane, upright spine, neutral gaze, and a minimum stride.
Standing calf raises
Tracks heel-lift, verifies real vertical lift, enforces locked knees, measures left-vs-right symmetry, and flags body sway.
From rehab-gentle to elite-strict.
The same engine adjusts from a physio’s clinic to an elite training floor. It’s a natural fit for our healthcare and AI/ML engineering work.
Home athletes
A second set of expert eyes when you train alone — so bad reps get caught instead of compounding.
Physiotherapists
Objective range-of-motion and symmetry tracking through rehab, tunable from gentle to strict per stage.
Elite coaches
Frame-accurate biomechanics and reviewable session data to fine-tune athletic performance.
Your idea, running in days.
VisionFiT started as a proof of concept. Tell us the movement, the metric, or the workflow you want to see, and we'll stand up a working build on your data — fast.