Internal technical demonstration
PosePilot — On-Device AI Photography Coach
PosePilot is an iOS app that acts as an AI photography coach — its guiding idea is "AI helps you shoot, not fix": it coaches pose, composition and lighting in real time instead of editing the photo afterward.
Technology stack
- SwiftUI
- XCTest
- Fastify
- TypeScript
- Prisma
- PostgreSQL
- Docker Compose
- Nginx
Challenge
People taking selfies or solo shots want the kind of real-time guidance a photographer would give, but without sending camera frames to a server or judging pose, framing and lighting by themselves.
Approach
On-device skeleton detection compares the live camera feed against a library of 20+ poses across ten scenarios (business, street, travel, fitness, and more), scores composition against rule-of-thirds, headroom and subject placement, and checks lighting — the shutter fires automatically once all three clear threshold.
Architecture
A SwiftUI client owns capture and on-device pose/composition scoring; a Fastify + TypeScript + Prisma + PostgreSQL backend handles accounts and subscriptions; the backend is deployed with Docker Compose behind Nginx. Camera frames never leave the device.
Outcome
Shipped on the App Store with six-language localization (English, Chinese, Japanese, Korean, Spanish, French), a free tier plus monthly/annual Pro subscription, and an iOS 17+ requirement for the on-device vision workload.