Role
An internal AI service for face-recognition attendance in Core CMS (mobile app and in-store kiosk). It runs as its own microservice so heavy inference never affects the business API.
Pipeline
- Face detection — SCRFD
- Tracking across frames — ByteTrack
- Anti-spoofing (printed photos, screens) — MiniFASNet
- Feature extraction — ArcFace (MobileFaceNet, 512-dimensional vectors)
- 1:1 verification against the employee's enrolled face
Everything runs on CPU with ONNX Runtime — no GPU needed, a good fit for self-hosted infrastructure.
Integration
- API: verification, embedding extraction, tracking, WebSocket streaming, image quality checks at enrollment, health checks.
- Employee logic, face enrollment and attendance live in Core CMS; embeddings are stored in PostgreSQL (pgvector).
- Runs only inside the cluster (not public), with Prometheus metrics.