Personal
Why I Built PuffinVBT
The personal story behind PuffinVBT. A decade of lifting, a stubborn ego, coaches who know my body better than I do, and the realization that bar speed doesn't lie.
PuffinVBT
Smartphone-based Velocity Tracking for Powerlifters
PuffinVBT turns ordinary training videos into detailed barbell velocity profiles using computer vision - no special equipment required. Built on Master's thesis research (see below).
Get StartedFree to start - 15 analyses, no card required. See pricing.
How It Works
1. Record
Film your set with any smartphone. Side view works best.
2. Upload
Upload the video with exercise type, weight, reps, and RPE.
3. AI Analysis
Machine learning segmentation + ellipse fitting extract barbell velocity frame-by-frame.
4. Insights
Per-rep velocity, ROM, phase timing, trends, and coach feedback.
Platform Features
VBT Analysis
Peak & mean velocity, ROM, concentric/eccentric split, velocity decay, and rep consistency - all computed automatically from video.
Dashboard
At-a-glance summary of sessions, exercises, average RPE, and recent uploads.
Trends & Progression
e1RM progression, load-velocity profiles, readiness tracking, rep consistency, personal records, and auto-generated insights.
Athlete Profile
Track bodyweight over time, log competition results, set goals, and calculate Dots & IPF GL points.
Coaching
Coaches invite athletes, view their videos, trends, and profile. Approval-based linking ensures privacy.
Feedback & Messaging
Threaded comments on individual videos plus direct messaging between coach and athlete.
Notifications
Real-time alerts for coach requests, new uploads, feedback, messages, and completed analyses.
Dark Mode
Full light/dark theme toggle, respects system preference.
Privacy First
Google SSO login, data ownership, GDPR-aware design, and opt-out from research use.
Research Background
This platform is the practical component of a Master's thesis investigating two research questions:
- RQ1 - How accurately can smartphone video + machine learning segmentation + ellipse fitting measure barbell velocity and range of motion compared to a tethered sensor (RepOne)?
- RQ2 - Can classical ML models trained on VBT metrics predict athlete-reported RPE?
Built with Flask, Bootstrap, Chart.js, Google Cloud Run, and a custom 6-step ML pipeline (preprocess → segment → ellipse fit → cleanse → perspective transform → rep detection).
Get Started
What you provide
- Training videos (squat, bench, deadlift)
- RPE ratings and session details
- Basic lift info (weight, reps, variation)
What you get back
- Per-rep velocity & ROM breakdowns
- Progression tracking & personal records
- Coach collaboration tools
- Bodyweight & strength score tracking
15 analyses free, no card required. Athlete and Coach plans unlock unlimited analyses after that - see pricing. Sharing training data for the underlying research is optional and asked for separately during sign-up consent, never a requirement to use the product.
From the blog
All posts →