Zepto Open-Sources AI Face Authentication Platform Built for India’s Dark Stores

Quick commerce company Zepto has open-sourced its in-house face authentication platform, ZepIris, a system developed to manage employee attendance and identity verification across its network of dark stores and delivery operations.

The announcement was made by Nikhil Mittal, Zepto Chief Technology Officer, who said the company built the platform after existing commercial solutions failed to perform reliably in real-world operating conditions such as low-light warehouses, budget smartphones and large-scale workforce deployments.

According to Mittal, the platform is already processing hundreds of thousands of face verifications daily across Zepto’s operations.

The system is designed to eliminate traditional attendance methods such as OTPs, physical registers and manual verification processes while preventing “buddy punching,” where employees mark attendance on behalf of others.

ZepIris combines multiple AI and machine learning technologies, including on-device face detection through Google ML Kit, ArcFace embeddings using InsightFace, vector search powered by Milvus and custom classifiers built on MobileNet and ResNet architectures.

The platform also includes spoof, blur and nudity detection systems running simultaneously to maintain low latency during authentication.

“We're open-sourcing it because the problem isn't unique to us,” Mittal wrote in a LinkedIn post. “Any team running identity verification at scale — attendance, onboarding, access control — deserves infrastructure that actually works in the real world, not just in controlled demos.”