Facenex

The Problem

Carrying cards is friction. Forgetting your wallet kills the transaction. Contactless payments solved speed but not identity — anyone with the card can pay. We wanted to eliminate the card entirely while making payments more secure, not less.

What We Built

Facenex is a facial recognition payment system. The end user walks up to a merchant's PoS terminal, looks at the camera, and enters a PIN. The system authenticates them via facial biometrics and authorises the charge. No card. No wallet. No phone.

How It Works

1. Enrollment — users register their face and set a PIN via a mobile app. Face templates are stored as encrypted vector embeddings, never as raw images.
2. Transaction — at the PoS, the merchant enters the amount. The user looks at the camera and enters their PIN.
3. Authentication — the system extracts a face embedding from the live camera feed, matches it against the enrolled template, and verifies the PIN. Two-factor: something you are (face) + something you know (PIN).
4. Authorisation — on successful authentication, the charge is processed through the payment gateway. The entire flow takes under 3 seconds.

Architecture

• Biometric pipeline — face detection (Haar cascades / SSD), alignment, embedding extraction (deep neural network), and vector matching (cosine similarity against enrolled templates).
• Liveness detection — the system checks for blink rates, head micro-movements, and depth cues to prevent photo/video spoofing.
• PIN hardening — PINs are hashed with bcrypt and salted. Rate limiting prevents brute force. After 3 failed attempts, the account locks.
• PoS integration — the terminal software runs on Android with a React Native frontend. It communicates with the backend via encrypted WebSocket for real-time authentication.
• Payment gateway — Facenex integrates with existing payment processors. It handles authentication; the gateway handles the money movement.

What This Proves

Biometric payments require solving three hard problems simultaneously: accuracy (false accept/reject rates), security (anti-spoofing), and UX (speed at the counter). We solved all three. The system was designed to process transactions in under 3 seconds with a false accept rate below 0.001%.

→ Engineering deep dive: Face recognition payments at the point of sale — architecture, security, and UX