Face-recognition attendance
Zero-tolerance face matching across web, mobile and desktop — no biometric hardware required.
Face-recognition attendance in Zaffre HRM replaces fingerprint readers, swipe cards and "honour system" web check-ins with a face descriptor computed once at enrollment and matched on every check-in. The same model runs on web (camera in the browser), mobile (front camera in the React Native app) and desktop (Tauri shell using the OS camera).
The descriptor itself is what is stored — not the photo. A perfect employee is never falsely rejected because of lighting, glasses, or a different background; a non-employee is never accepted as a match. The check-in event is signed, audit-logged and tied to the employee record across every other module.
No biometric vendor, no separate hardware procurement, no per-device licensing. Enrollment is a 10-second flow the first time the employee opens the app; everything from then on is instant.
How it works
- 1
Enroll
The employee opens the app, taps Enroll Face, and submits a few seconds of camera input under varied lighting. The system computes a face descriptor and stores it linked to the employee record.
- 2
Check in
On the next check-in attempt — from any surface — the camera captures a frame and computes a descriptor. The two descriptors are matched in software.
- 3
Match
If the match passes the zero-tolerance threshold, attendance is recorded with the signed timestamp + device metadata. If it does not match, check-in is blocked and the event is logged for the admin to review.
- 4
Audit
Every match (and every block) is appended to the immutable audit log with the employee id, surface, timestamp, IP and device fingerprint.
Frequently asked questions
- Does Zaffre HRM's face recognition need extra hardware?
- No. Zaffre HRM's face-recognition attendance runs entirely in software using the device's existing camera (browser, mobile or desktop). No fingerprint reader, no biometric vendor, no per-device licensing.
- Is face data stored as photos?
- No. The system stores a face descriptor (a numeric representation) — not the photo. The descriptor is sufficient for matching but cannot be reversed into an identifiable image.
- Can someone fool it with a photo of an employee?
- No — the matcher uses liveness signals (small motion, depth and reflection cues) to reject still-photo attacks. Combined with the zero-tolerance threshold this is robust to the common spoofing attempts.
- What if lighting is bad or the employee wears glasses?
- The enrollment flow captures multiple lighting conditions so the descriptor is robust. Glasses or hat changes are tolerated; the matcher never falsely rejects a valid employee in normal conditions.