• 

  Merchant POS with Nova risk scoring + voice output

• 

  Amazon Nova 3-model architecture breakdown

• 

  Landing page

• 

  Distress Mode with SOS SMS + GPS alert

Inspiration

India has over 400 million unbanked citizens who can't access digital payments because they don't own a smartphone. Even for smartphone users, emergencies happen — your phone dies, gets stolen, or you simply forgot it. We asked: what if your body was your wallet? That question led to PulsePay.

What it does

PulsePay is a biometric micro-wallet that lets users pay using only their face, fingerprint, and voice — no phone required.

• Adaptive Authentication — Amazon Nova 2 Lite dynamically scores transaction risk and decides whether a face scan alone is enough, or if fingerprint verification is also needed
• Zero-Knowledge Biometrics — Face images are never stored. Amazon Nova Multimodal Embeddings converts facial features into mathematical vectors, which are then SHA3-256 hashed
• Voice-Confirmed Payments — Amazon Nova 2 Sonic provides real-time spoken confirmations so merchants know the payment went through
• Distress Mode — Scan your pinky finger instead of your index finger to silently trigger an SOS alert with GPS coordinates to your family
• 72-Hour Auto-Expiry — Wallets self-destruct after 72 hours with automatic refunds
• Family Wallet Delegation — Add family members with their own biometrics and spending caps
• UPI & Crypto Wallet Connect — Link existing UPI handles or crypto wallets for funding

How we built it

• Frontend: Next.js 16 + TypeScript + Tailwind CSS as a PWA with glassmorphism UI
• Backend: Node.js + Express integrating three Amazon Nova models via AWS Bedrock SDK
• CV Service: Python FastAPI handling biometric processing with Nova Multimodal Embeddings
• Database: PostgreSQL for wallet state, transactions, and family delegation

Amazon Nova Integration:

1. Nova Multimodal Embeddings — Generates high-dimensional face vectors for zero-knowledge biometric matching
2. Nova 2 Lite — Agentic security layer analyzing transaction context to dynamically set authentication requirements
3. Nova 2 Sonic — Bidirectional streaming voice AI generating natural merchant confirmations in real-time

Challenges we ran into

• Bidirectional streaming with Nova 2 Sonic required deep research into the AWS SDK's InvokeModelWithBidirectionalStreamCommand and NodeHttp2Handler — the event protocol was complex to implement correctly
• Balancing security with usability — the AI risk engine needed to be smart enough to not annoy users on small purchases while catching suspicious transactions
• Camera-based fingerprint scanning in a browser was unreliable, so we engineered a touch-based fingerprint solution with visual progress feedback

Accomplishments that we're proud of

• Built a complete biometric payment system that works without a phone
• Integrated three different Amazon Nova models across three hackathon categories (Multimodal, Agentic, Voice)
• The Distress Mode feature could genuinely save lives in real-world emergencies
• Achieved zero-knowledge biometric storage — no raw biometric data is ever persisted

What we learned

• Amazon Nova's multimodal capabilities can replace traditional specialized ML models for biometric embedding
• Nova 2 Lite effectively acts as an autonomous security agent when given the right context
• Building accessibility-first (no phone, no literacy required) forces better design decisions for everyone

What's next for PulsePay

• Integration with India's UIDAI Aadhaar system for real identity verification
• Hardware biometric sensor support for enterprise merchant terminals
• Expanding Nova 2 Sonic to support Hindi, Tamil, and other regional languages
• Pilot program with rural merchants in partnership with local banks

Built With

• amazon-nova-2-lite
• amazon-nova-2-sonic
• amazon-nova-multimodal-embeddings
• audio
• aws-bedrock
• express.js
• fastapi
• next.js-16
• node.js
• postgresql
• python
• tailwind-css
• typescript
• web
• web-crypto-api