Home page
Firebase auth
Admin panel
Scanner
Scanner results
Scanner results final
Architecture diagram

CashGuard - AI-Powered Currency Authentication

💡 Inspiration

Counterfeit currency remains a persistent problem worldwide, affecting businesses and individuals alike. We wanted to create an accessible tool that leverages modern AI to help anyone quickly verify bill authenticity using just their smartphone camera.

🎯 What it does

CashGuard is a mobile-first web application that:

Scans currency bills using your device camera
Validates authenticity using dual AI models: Google Gemini Vision API for detailed security feature analysis, plus a custom Vertex AI classifier for real/fake detection
Provides instant feedback with confidence scores and detected security features
Stores scan history securely in Firebase with user-specific access control
Tracks analytics to understand usage patterns and improve accuracy

🛠️ How we built it

Frontend: Built with Next.js 14 (App Router) and React 18, styled with Tailwind CSS for a modern, responsive UI. We implemented camera access with automatic image compression to optimize performance.

AI Integration:

Connected to Google AI Studio's Gemini API for vision-based currency analysis
Trained a custom bill classifier using Vertex AI for binary real/fake classification
Combined both models for higher accuracy and detailed insights

Backend: Leveraged Firebase for authentication (Google Sign-In), Firestore for scan history, Storage for images, and Analytics/Performance monitoring.

Deployment: Containerized with Docker and deployed to Google Cloud Run for serverless, auto-scaling infrastructure.

📚 What we learned

How to optimize mobile camera workflows for real-time image capture
Prompt engineering techniques for Gemini Vision API to extract specific security features
Implementing Firebase security rules to protect user data
Cloud Run deployment patterns for Next.js applications
Rate limiting and API key security best practices

🚧 Challenges we faced

Mobile camera optimization: Getting reliable camera access across different browsers and devices required extensive testing
Firebase auth domains: Cloud Run deployment initially broke authentication; solved by programmatically adding authorized domains
Image processing: Balancing compression quality vs. file size for fast uploads without losing security feature details
Rate limiting: Implemented proper throttling to prevent API abuse while maintaining good UX