Skip to content

Behavior DNA

05.2026

Behavior DNA

Continuous behavioral authentication for digital banking — verifying who is using an account, not just which credentials they hold.

ReactTypeScriptFastAPIXGBoostSHAPHyperledger FabricPostgreSQLRedisDocker
GitHub ↗

Built for the CBI Hackathon. Passwords prove what you know and OTPs prove what you have — Behavior DNA proves who you are.

The idea

Stolen credentials and SIM-swaps pass every traditional check. Behavior DNA watches how a user types and touches from the first keypress, so an attacker holding the right password still looks wrong.

How it works

  • An on-device SDK captures 40+ keystroke, touch and session signals and flushes them every 500 ms. Device identifiers are HMAC-hashed on the device and never leave in the clear.
  • A FastAPI decision engine evaluates every login through three parallel pathways: behavioral machine learning, blockchain-anchored device trust, and coercion context.
  • Each login returns APPROVE / STEP_UP / HARD_BLOCK in under 80 ms, with the top SHAP attributions behind every decision.

Security features

  • SIM-swap defense — a device-hash baseline on Hyperledger Fabric; a mismatch is blocked outright.
  • Duress protection — a silent duress PIN opens a decoy session and alerts trusted contacts.
  • Coercion detection — screen-sharing, hesitation and paste patterns raise the risk score.
  • Post-quantum crypto — ML-KEM-768 key encapsulation and ML-DSA-65 signatures (FIPS 203/204).
  • Active deception — honeypot fields and decoy endpoints that fingerprint intruders.

Stack

React · TypeScript · FastAPI · XGBoost · SHAP · Hyperledger Fabric · PostgreSQL · Redis · Docker

Generating world

Loading level · 0%Tip: press Esc on any screen to go backClick or press any key to skip