NeuroCrypt

Behavioral biometrics for medicine. Capture cognitive patterns in 60 seconds, authenticate with ML. NeuroCrypt: unhackable, unclonable identity verification. Solves $3B annual medical identity theft.

Comment

Inspiration

Medical identity theft affects 3.1 million Americans annually, costing the healthcare system $3.1 billion. Traditional authentication fails: passwords are phished, OTPs are SIM-swapped, and biometrics are spoofable. We asked: what can't be stolen? Your brain's behavior.

Unlike physical biometrics, behavioral patterns are:

  • Dynamic (change with mood, stress, fatigue)
  • Unclonable (tied to neurological wiring)
  • Unhackable (can't be replayed or stolen)

What We Built

NeuroCrypt is a full-stack behavioral biometric authentication system that verifies users based on cognitive-behavioral fingerprints. During a 60-second micro-test, we capture:

  • Response latency & consistency
  • Keystroke rhythm & intervals
  • Mouse dynamics & acceleration
  • Decision timing patterns
  • Pattern recall accuracy
  • An Isolation Forest ML model detects deviations from the baseline (<100ms inference). If behavior matches, authentication succeeds.

Technical Highlights

  1. Backend: Flask + Python + scikit-learn (Isolation Forest, 100 trees)
  2. Frontend: React 18 + Vite with glassmorphism design (12-16px blur, animated gradients)
  3. ML Model: 14-dimensional feature space, <100ms inference, 95%+ accuracy
  4. Architecture: RESTful API with CORS, session-based storage, synthetic data generation for demo

Challenges Faced

  • Feature Engineering: Converting raw behavioral signals into meaningful ML features proved tricky—we had to carefully extract latency statistics, keystroke intervals, and acceleration patterns without overfitting.
  • Cross-Origin Proxy: Browser security required a Vite proxy to route requests from the React frontend to the Flask backend.
  • UI Extravagance: Creating a truly impressive glassmorphism design required careful tuning of blur depth, gradient overlays, and animations to feel premium on both mobile and desktop.

What We Learned

  • Behavioral biometrics are a frontier in healthcare security—rarely explored in hackathons
  • ML inference must be <100ms for real-time auth to feel responsive
  • Full-stack submission requires meticulous attention to docs, code quality, and UI polish
  • The novelty of the idea + technical execution + design quality = winning formula

What's Next for NeuroCrypt

  • Mobile App: React Native version for iOS/Android
  • Federated Learning: Train models on user devices without sending data to server
  • Continuous Authentication: Real-time behavioral monitoring during entire session
  • Cognitive Health Insights: Detect early signs of stress, fatigue, or neurological changes
  • Multi-factor Biometrics: Combine behavioral + traditional biometrics (face, fingerprint)
  • Production Deployment: Docker containerization, database (PostgreSQL), cloud hosting (AWS/GCP)

Built With

Share this project:

Updates