On-device face recognition for encrypted photo storage

Consumer Privacy

Industry: Consumer Privacy

Platform: iOS

Stack: Swift · CoreML · CryptoKit · Vision · Accelerate

Description: On-device face-recognition pipeline feeding an AES-256-GCM vault with resilient self-destruct.

---

The engineering surface was a zero-network face-search and storage pipeline on iOS. The recognition layer combined Vision face detection with a CoreML embedding pipeline tuned for mobile memory pressure. The storage layer was hardware-backed AES-256-GCM with atomic two-phase writes and a self-destruct state machine that survives force-quit and reboot.

We owned the full product build — recognition pipeline, vault, key management, self-destruct state machine, and StoreKit entitlement layer. The client owned the brand, paywall content, and App Store relationship.

Engagement: 4-month build · 2-engineer team · 2025.

Engineering signal: sub-second per-photo end-to-end on recent iPhone hardware; zero data loss across simulated crash scenarios.