360 Facevision HRMS

The problem

Face-recognition attendance is trivially defeated by a photo on a phone screen. Without liveness detection, biometric HRMS creates a fraud surface rather than closing one.

The approach

Built a face recognition HRMS with anti-spoofing and emotion detection. ResNet-50 and ArcFace handle recognition and embedding, MobileNetV3 keeps the liveness check fast enough to run inline, Redis caches embeddings for low-latency matching, and Django with React serves the HR workflows and AI-driven payroll rules.

Results

• Anti-spoof attendance running in production.
• AI payroll rules driven directly from verified attendance data.
• Emotion detection layered onto the same recognition pipeline.

Engineering notes

MobileNetV3 for the anti-spoofing pass keeps the liveness check cheap enough to run on every attendance event rather than sampling.