Powor
LiveGym Community · 2025 — Present
What Strava did for running, for lifting. A social feed, following, PRs and rankings layered on a full training tracker. Every layer is mine — the app, the API, the ML behind Body Scan, and the release pipeline.
Role: Founder & Lead Engineer — solo build
- Users
- 600+Users
- Conversion lift
- 30%+Conversion lift
- Languages shipped
- 5Languages shipped
- React Native
- Expo SDK 57
- NestJS
- MongoDB Atlas
- watchOS
- SwiftUI
- CoreML
- RevenueCat
- OpenAI API
- Xcode Cloud
Engineering breakdown
The problem
Most training apps are solitary logbooks, and people quit them. Consistency comes from progress being visible to the people you train alongside — the thing that made Strava work for runners, and that lifting never really got.
What I built
- Built the social layer end to end — feed, following, likes and comments, rankings and challenges — so a logged session becomes something friends actually see.
- Modelled the training domain properly: sets, reps, load and RPE per exercise, feeding PR detection, progression analytics, streaks and achievement badges through MongoDB aggregation pipelines rather than client-side maths.
- Shipped Body Scan on a dual-tier OpenAI vision pipeline, routing the common path to a cheaper model and escalating only on low confidence, so per-request cost stays flat as usage grows.
- Architected IMU-based exercise recognition with CoreML, running inference on-device so logging keeps working without a network round trip.
- Engineered A/B-tested RevenueCat paywalls, free-trial flows and multi-currency pricing that lifted subscription conversion by 30%+.
- Localised into five languages and set up Xcode Cloud and EAS Build CI/CD with TestFlight distribution — v1.17 shipped since the November 2025 launch, zero rejected builds.



