AI / Computer VisionCase Study

How we replaced badge swiping with AI face recognition — for 500+ employees.

A manufacturing facility in Chhattisgarh needed a faster, fraud-proof attendance system. We built one that recognizes faces in under a second — with zero cloud dependency.

Spring BootPostgreSQLpgvectorPythonFaceNetAndroid ML KitDocker

The Problem

Badge swiping was slow, fraud-prone, and hated by everyone.

The facility had 500+ workers across multiple shifts. Every morning, a 20-minute queue formed at the badge readers. Buddy punching was rampant — workers swiped for absent colleagues, costing the company thousands in ghost payroll.

They tried fingerprint scanners. Workers with calloused hands from the factory floor had a 30% failure rate. The system was abandoned within weeks.

They needed something that was fast, fraud-proof, and worked for every hand — literally.

Our Approach

Face-first. Cloud-free. Sub-second.

On-device face detection

Android tablets at each entry point run ML Kit for real-time face detection. The face is captured and sent to the server — no cloud round-trips, no internet dependency. Works even during network outages with a local queue.

FaceNet embeddings + pgvector similarity search

Each face is converted to a 512-dimensional embedding using FaceNet/ArcFace. These embeddings are stored in PostgreSQL with pgvector, enabling cosine similarity search across all 500+ employees in under 50ms.

Production-hardened from day one

Dockerized deployment, health checks, automated backups, and a real-time dashboard showing live attendance, late arrivals, and shift analytics. The operations team sees everything — no more manual reconciliation.

Timeline

From kickoff to production in 10 weeks.

Week 1–2

Discovery & Architecture

Stakeholder interviews, system design, face-encoding pipeline design

Week 3–5

Core Engine

FaceNet/ArcFace integration, pgvector similarity search, Spring Boot APIs

Week 5–7

Android App

On-device ML Kit face detection, offline queue, camera optimization

Week 7–8

Dashboard & Reports

Real-time attendance dashboard, Excel exports, admin controls

Week 9–10

Deployment & Hardening

Docker deployment, load testing, monitoring, security audit

The Results

Numbers that speak for themselves.

500+Employees processed daily
<1sFace match latency
99.7%Recognition accuracy
0Cloud dependency — fully on-premise
3xFaster than badge swiping
24/7System uptime with monitoring
developersEra didn't just build our attendance system — they engineered something that handles 500+ employees daily without a single hiccup. The face recognition is faster than badge swiping. We haven't had a single buddy-punching incident since.

Rajesh M.

Operations Director, Manufacturing Facility, Chhattisgarh

Have a similar challenge?
Let's solve it together.

Whether it's AI, automation, or a system that needs to handle thousands of users — we'll scope it for free.

Start a Conversation →