Face Attendance System

Face-verified clock-in on phone and door terminals.

Role: Product design · Full stack
Year: 2026
Platform: Web dashboard · iOS · Android · Face terminals
Stack: Laravel 12 · React 19 · Expo · DeepFace · MariaDB · Docker

Overview

The problem

A public health office tracked attendance across several clinics with a mix of door terminals and manual sign-ins. Punches were duplicated or missing, field staff couldn't clock in at all, and HR had no quick way to tell a genuine punch from a doubtful one.

The idea

One attendance pipeline for every capture method: staff clock in with a selfie and GPS on their phone or at a face terminal, each punch is checked against their enrolled face, and only the doubtful ones land in an HR review queue.

Built for a public health office. Shown anonymised — screens are recreated with sample data.

What I did

Design

01Live dashboard: who's in, late, not in yet and what needs triage, per site
02Review queue that sorts mismatches first so HR clears the worst cases quickly
03Mobile clock-in with a guided face frame, head-turn liveness prompts and location status
04Per-office capture policy settings written in plain language, not config flags

Engineering

01Laravel 12 API with a strict Form Request → DTO → Service layering and separate web and mobile APIs
02Self-hosted DeepFace (Facenet512) service: embeddings are compared in PHP, verdicts sorted into verified / review / unverified
03Door-terminal push ingest and phone punches share one dedup and in/out pairing pipeline, with row locks and replay-safe indexes
04Multi-office tenancy with per-office capture policy, haversine geofence and per-employee exemptions