SuperGrid

The problem

A large construction and interior design firm in Delhi was facing widespread attendance fraud, employees claiming to be on-site while physically absent. With 1000+ employees spread across multiple sites, each checking in multiple times daily, manual registers and traditional methods were completely inadequate. There was no reliable way to verify presence, leading to payroll fraud, accountability gaps, and operational inefficiency.

Our solution

We built SuperGrid, a production-grade face recognition attendance system with geofence validation and GPS integration. Every check-in requires a real-time face match against a stored 128-dimensional embedding, combined with geofence verification to confirm the employee is physically within the designated site boundary. The system handles 100-200 requests per second across all sites simultaneously.

What we built

• Face enrollment with 128-dim embedding generation and storage
• Real-time face verification with liveness detection and strict 0.4 threshold matching
• Geofence-based attendance validation via GPS/Traccar integration
• Device management dashboard for monitoring GPS devices across all sites
• OTP-based supervisor authentication via Twilio SMS for admin overrides
• Redis distributed queue and Celery background workers for high concurrency
• PostgreSQL with connection pooling (100 base + 150 overflow) for performance at scale
• Full device management and supervisor bypass interface