2024–2025·Android · IA on-device · ML Kit · TensorFlow Lite · Visão Computacional
Maia · Facial Attendance Control
On-device facial biometrics at 10 FPS for corporate events
Context
Attendance control at corporate events historically suffers from queues, lost badges, smudged QR codes. Maia's pitch: replace all that with facial biometrics running directly on the staff's Android — no dependency on stable internet.
Architecture
- Clean Architecture + MVVM with Hilt for DI — modular and testable code.
- UI 100% Jetpack Compose with Material Design 3.
- CameraX capture with frame processing, standardized 112×112px cropping and optimized multipart upload to the AWS backend.
- AI pipeline using Google ML Kit + TensorFlow Lite, reaching 10 FPS with adaptive AOI validation.
- Resilience via Sentry, smart retry logic and tracking IDs to guarantee data integrity on unstable networks.
Outcome
Solution in production, replacing legacy methods with sub-second latency and offline-first operation.