โ† Back to work
08[Archived]

Banklytics

AI-powered fraud detection system.

[ Visit project โ†’ ]

Summary

An AI fraud-detection system for banks that replaces SMS one-time codes with real-time voice verification and ML-based transaction risk scoring.

Problem you're solving

SMS-based two-factor authentication is trivially bypassed by SIM swaps, phishing, and now AI voice-cloned scam calls, yet it's still the default fraud check at most banks.

Target user

Banks and fintechs that need stronger transaction verification, plus their end users โ€” especially those without a smartphone or biometric hardware to fall back on.

Competitors

Legacy SMS 2FA is the incumbent being replaced; app-based biometric auth from banks and providers like Plaid excludes users without a smartphone.

Insight

The fraud problem isn't just detecting risky transactions โ€” it's verifying identity in a way AI voice cloning can't trivially defeat, and that verification shouldn't require a smartphone.

Solution

A serverless pipeline scores transactions in real time with an ML model, then confirms high-risk transactions via an AI-driven phone call using voice verification, with an optional facial-recognition layer for app users and a dashboard for fraud analysts.

  • Next.js
  • Tailwind CSS
  • OpenAI
  • RetellAI
  • Supabase

Distribution

Built in a single hackathon weekend; next steps identified were partnering with fintechs for API integration and deploying a live prototype.

Adapting to user feedback

The core design tension โ€” balancing millisecond-speed verification calls against the risk of AI-cloned voices defeating the system โ€” was flagged as the top priority for the next iteration, ahead of voice liveness detection.