behavioral authentication system

Multi-Agent Authentication System

Comment

Behavioral Authentication System

Inspiration

Traditional authentication systems rely heavily on passwords or OTPs, which are vulnerable to phishing, brute-force attacks, and user fatigue. We were inspired by how human behavior—like typing patterns, swipe gestures, and usage rhythm—can uniquely identify individuals. Our goal was to explore a privacy-conscious, frictionless authentication method powered by a multi-agent AI system that learns and adapts to behavioral patterns in real time.

What it does

Our system performs real-time user authentication based on behavioral data (e.g., typing speed, interaction sequences) using a Multi-Agent Architecture:

  • Each agent analyzes a specific behavior trait (e.g., motion, input speed, device usage).
  • A shared Siamese GRU model compares current behavior with historical patterns.
  • Agents communicate via Pub/Sub, coordinate responses, and make a collective authentication decision.
  • The frontend (built with Flutter) allows users to log in seamlessly without passwords—just by being themselves.

How we built it

  • Model: Siamese GRU trained on user behavior sequences.
  • Backend Agents: Lightweight Python agents running on GCP Compute Engine (f1-micro) to stay within free tier limits.
  • Communication: Google Cloud Pub/Sub used for agent coordination.
  • Storage: Firestore for user profiles and state; Cloud Storage for models.
  • API Layer: Cloud Functions for authentication endpoints.
  • Frontend: Flutter app with Firebase hosting, showcasing live behavioral authentication.
  • Analytics: BigQuery for behavioral trend analysis.

Challenges we ran into

  • Resource limits: Designing around GCP’s free tier (memory, compute, storage).
  • Model sharing: Ensuring shared use of a single GRU model across multiple agents efficiently.
  • Latency: Keeping inference and communication under 500ms for a smooth user experience.
  • Data privacy: Designing a system that doesn’t store raw input data but still learns behavior patterns.
  • Concurrency: Synchronizing agent decisions without overwhelming Firestore or Pub/Sub quotas.

Accomplishments that we're proud of

  • Successfully deployed a real-time AI-based authentication system on GCP's free tier.
  • Designed a fully modular and scalable multi-agent system.
  • Delivered a working Flutter app demonstrating passwordless login via behavioral data.
  • Kept the entire infrastructure cost at zero, ideal for startups or research prototypes.
  • Built an analytics dashboard using BigQuery to visualize behavioral trends.

What we learned

  • How to design microservices with shared state and models efficiently.
  • Real-world application of Pub/Sub-based event-driven architectures.
  • Importance of free tier optimization for startups and prototypes.
  • Strengths and limitations of behavioral biometrics in authentication.
  • How multi-agent systems can improve scalability and fault tolerance.

What's next for behavioral authentication system

  • Expand behavioral inputs: Integrate additional modalities like mouse dynamics or gyroscope data.
  • Federated learning: Implement client-side training for enhanced privacy.
  • Risk-based authentication: Adjust thresholds dynamically based on location, time, and device.
  • Cross-platform support: Extend to iOS, desktop, and browser plugins.
  • Developer SDK: Allow third-party developers to plug behavioral auth into their apps easily.
  • Security hardening: Introduce anomaly detection to prevent spoofing or mimicry attacks.

Built With

Share this project:

Updates