Real-Time Form Checking in Your Browser — No App Needed
By Ollie
--- title: "Real-Time Form Checking in Your Browser — No App Needed" description: "See AI analyse your squat, deadlift, and push-up form in real time. No download, no app, no data sent anywhere. Just your webcam and your browser." ---
Real-Time Form Checking in Your Browser — No App Needed
What if your browser could tell you when your squat depth is off, your knees are caving in, or your push-up form is breaking down?
We built it. And you can try it right now.
Try the Live Demo
No app to download. No account to create. No data sent anywhere. Just open the page, allow camera access, and start moving.
- Squat — depth, knee valgus, torso lean
- Deadlift — hip hinge angle, spine neutrality, knee position
- Push-up — depth at the elbow, body line alignment
- 🟢 Green — good form
- 🟡 Amber — needs attention
- 🔴 Red — form error detected
How It Works
The form checking uses MediaPipe Pose — Google's open-source pose estimation model. It detects 33 body landmarks from a single camera feed.
From those landmarks, we calculate joint angles in real time:
Squat depth: The angle at your knee (hip → knee → ankle). Below 90° is deep. Above 160° means you're barely bending.
Knee valgus: We compare the horizontal position of your knees vs. your ankles. If knees drift inward past a threshold, the system flags it.
Torso lean: The angle between your shoulders and hips relative to vertical. Excessive forward lean means you're compensating — usually a mobility or core stability issue.
The same approach works for deadlift hinge angles and push-up body alignment.
Why Browser-Side Matters
Every existing "AI trainer" app processes your video on their servers. Your workout footage gets uploaded, processed, and stored somewhere you can't see.
PumplAI's form checking runs entirely in your browser. The AI model loads once, then processes every video frame locally. No video data is ever sent to any server.
This matters for three reasons:
1. Privacy. Your workout footage is nobody's business but yours. 2. Speed. No upload latency. Analysis happens at 30+ FPS in real time. 3. GDPR. For EU users, there's no data transfer to worry about. The processing happens on your device.
What Trainers Can Do With This
For personal trainers, browser-side form checking opens up async coaching at scale:
- Client self-checks between sessions. Clients can verify their form at home without scheduling a video call.
- Technique cues that stick. Real-time feedback reinforces coaching cues more effectively than delayed video review.
- Scalable quality. You can coach more clients without sacrificing form oversight.
This isn't a replacement for a skilled trainer's eye. It's a supplement — catching the obvious errors so trainers can focus on the nuanced coaching that actually requires expertise.
The Tech Stack
For the technically curious:
- MediaPipe Pose Landmarker (Lite model, float16) — 33 landmarks per frame
- WebGL GPU delegate — hardware-accelerated inference in the browser
- Canvas overlay — real-time skeleton drawing with colour-coded feedback
- Angle calculation — atan2-based joint angle computation for each exercise
- React + Next.js — server-side rendered page shell with client-side MediaPipe
The entire demo is under 200 lines of TypeScript. No backend. No API calls. Just math and a camera.
What's Next
The current demo covers three exercises with basic angle thresholds. We're building towards:
- More exercises — lunges, overhead press, rows, planks
- Rep counting — automatic set and rep tracking from pose data
- Movement quality scoring — aggregate form scores across a workout
- Trainer-defined thresholds — coaches can set custom form standards per client
All still browser-side. All still private.
---
PumplAI combines AI-matched coaching with real-time form checking. No other platform offers this.