WinkID SDK
Multimodal biometric auth in production — face, palm and voice behind a single authenticate() call.
Product engineer — architecture, Next.js + React Native frontends, .NET SDK APIs
The problem
Wink Cloud's thesis is that identity should be biometric, not password-based. Turning that thesis into something a merchant can actually integrate meant building an SDK for face, palm, and voice recognition that a frontend engineer could drop into a checkout flow without needing to understand any of the underlying model orchestration.
The approach
Multi-modal biometric SDK with a clean JS/TS interface and dynamic .NET APIs behind it. Frontend is Next.js, using MediaPipe for face and palm capture in the browser, and VAD (voice activity detection) for the voice modality. The customer-facing e-commerce app is the reference implementation that proves the SDK works in a real transaction flow — biometric auth gates the payment step.
Selected details
- SDK exposes the three modalities behind a single
.authenticate()call with a fallback chain — face preferred, palm if lighting is poor, voice if palm fails. - All biometric capture happens client-side. Only encrypted templates leave the device, never raw images.
- The reference e-commerce app doubled as our first-party test bench for every SDK regression.
- Bundle-split the SDK so merchants only ship the modalities they actually use.
Outcome
Shipped as Wink's foundational production POC. Being extended into a merchant-facing branding preview system so merchants can theme the biometric auth UI to match their brand.
Stack