Guardian Nova: Global Scam Protection Agent

AI-powered scam detector that analyzes suspicious email and SMS screenshots using Amazon Nova Pro to provide instant risk assessments and security advice.

Comment

Inspiration

With the exponential rise in digital scams, phishing attacks, and fraudulent messages targeting people worldwide, we recognized an urgent need for an accessible tool that could help anyone—regardless of technical expertise—identify potential threats. Every day, millions of people receive suspicious emails and SMS messages that could lead to identity theft, financial loss, or compromised personal data. We were inspired to create Guardian Nova as a protective shield that leverages cutting-edge AI technology to democratize cybersecurity, making advanced scam detection available to everyone with just a screenshot.

What it does

Guardian Nova is an intelligent scam detection application that analyzes screenshots of suspicious emails and SMS messages using Amazon Nova Pro's multimodal AI capabilities. Users simply upload a screenshot of a questionable message, and within seconds, Guardian Nova provides:

Risk Level Classification - Categorizes threats as High, Medium, or Low with color-coded visual indicators (red, yellow, green)

Detailed Analysis - Extracts sender information, identifies suspicious links, and analyzes message tone and content

Expert Reasoning - Provides comprehensive explanations of why a message is flagged as potentially dangerous

Actionable Security Advice - Offers specific recommendations on how to respond to the threat

Real-time Processing - Delivers instant results through an intuitive web interface

The application acts as a personal cybersecurity expert, helping users make informed decisions about suspicious communications before they become victims of scams.

How we built it

We built Guardian Nova using a modern, streamlined technology stack focused on simplicity and effectiveness:

Architecture: Single-file Streamlit application for easy deployment and maintenance

AI Integration: Integrated Amazon Nova Pro v1.0 multimodal AI model through AWS Bedrock Runtime. Designed custom system prompts to guide the AI in cybersecurity analysis. Implemented robust API payload construction with base64 image encoding. Structured JSON response parsing for consistent risk assessments.

Frontend Development: Built an intuitive UI with Streamlit for seamless user experience. Created secure credential input with password-masked fields. Implemented file upload functionality supporting JPG and PNG formats. Designed color-coded risk visualization for instant threat recognition.

Backend Logic: Developed comprehensive error handling for AWS authentication and connection issues. Implemented credential validation to prevent unauthorized API calls. Created modular functions for payload construction, analysis orchestration, and result display. Built image format detection and conversion pipeline.

Testing & Quality Assurance: Wrote 44 comprehensive unit and integration tests achieving 99% code coverage. Implemented property-based testing approach for correctness validation. Created checkpoint verification system for incremental development validation. Tested all error scenarios including authentication failures, network issues, and malformed responses.

Development Process: Followed a spec-driven development approach with detailed requirements, design, and task documentation. Used component-based architecture for maintainability and testability. Implemented incremental checkpoints to ensure quality at each development stage.

Challenges we ran into

Multimodal API Integration Complexity - Understanding and implementing the correct payload structure for Amazon Nova Pro's multimodal API was initially challenging. We had to carefully study the API documentation to properly format image data with base64 encoding and structure the messages array correctly.

Response Parsing Variability - The AI model's response structure required careful parsing through nested JSON objects. We had to handle the output -> message -> content hierarchy and extract text content that itself contained JSON, requiring multiple parsing stages.

Error Handling Comprehensiveness - Designing user-friendly error messages for various failure scenarios (authentication errors, connection timeouts, JSON parsing failures) while maintaining security best practices was complex. We needed to provide helpful troubleshooting guidance without exposing sensitive system information.

Credential Security - Balancing ease of use with security requirements meant implementing proper credential masking, validation, and ensuring credentials were never stored or logged while still providing a smooth user experience.

Testing Multimodal Interactions - Creating comprehensive tests for image upload, conversion, and API interactions required extensive mocking strategies to simulate AWS Bedrock responses without making actual API calls during testing.

Single-File Architecture Constraints - Maintaining clean, organized code within a single-file architecture while implementing multiple components (UI, business logic, error handling, API integration) required careful function organization and documentation.

Accomplishments that we're proud of

Production-Ready Application - Built a fully functional, tested application with 44/44 tests passing and 99% code coverage

Intuitive User Experience - Created a clean, accessible interface that anyone can use without technical knowledge—just upload a screenshot and get instant results

Robust Error Handling - Implemented comprehensive error handling that gracefully manages authentication failures, network issues, and unexpected responses with helpful troubleshooting guidance

Security-First Design - Ensured AWS credentials are properly masked, validated, and never stored, while preventing unauthorized API calls

Real-World Impact - Developed a tool that can genuinely help protect people from scams and phishing attacks in their daily digital lives

Clean Architecture - Achieved a well-organized, maintainable codebase with modular functions and clear separation of concerns despite single-file constraints

Comprehensive Testing - Created a thorough test suite covering unit tests, integration tests, and edge cases, ensuring reliability and correctness

Multimodal AI Integration - Successfully integrated cutting-edge Amazon Nova Pro multimodal AI for sophisticated image analysis and threat detection

What we learned

Technical Learnings: Deep understanding of AWS Bedrock Runtime API and multimodal AI model integration. Advanced Streamlit techniques for building interactive web applications with secure input handling. Best practices for base64 encoding and image format detection in Python. Comprehensive error handling strategies for cloud service integrations. Property-based testing approaches for validating system correctness across input spaces.

AI/ML Insights: How to craft effective system prompts to guide AI models toward specific analysis tasks. The importance of structured output formats (JSON) for reliable AI response parsing. Multimodal AI capabilities in analyzing visual content for cybersecurity applications. Balancing AI model parameters (temperature, max_tokens) for consistent results.

Security Principles: Credential management best practices in web applications. The importance of input validation and sanitization. How to provide helpful error messages without exposing sensitive system information. Designing user-friendly security tools that don't compromise on protection.

Development Process: The value of spec-driven development with clear requirements, design, and task documentation. Importance of incremental checkpoints for maintaining quality throughout development. How comprehensive testing catches edge cases and improves reliability. The benefits of modular architecture even within single-file constraints.

User Experience Design: The power of color-coded visual feedback for instant comprehension. Importance of providing both high-level results and detailed explanations. How to make advanced AI technology accessible to non-technical users.

What's next for Guardian Nova: Global Scam Protection Agent

Enhanced AI Capabilities: Support for analyzing multiple screenshots in batch mode. Integration with additional AI models for cross-validation and improved accuracy. Real-time URL scanning and domain reputation checking. Historical analysis to identify evolving scam patterns and trends.

Expanded Detection Features: Support for analyzing forwarded email headers and metadata. Voice message and voicemail transcript analysis. Social media post and message analysis. QR code scanning and validation.

User Experience Improvements: Browser extension for one-click screenshot analysis. Mobile application for on-the-go scam detection. Shareable analysis reports for warning others. Community-driven scam database with anonymized threat intelligence.

Enterprise Features: API access for integration with email security systems. Bulk analysis capabilities for organizations. Custom model training on organization-specific threats. Admin dashboard with analytics and reporting.

Educational Components: Interactive tutorials on identifying scam indicators. Scam awareness training modules. Regular threat intelligence updates and newsletters. Gamified learning experiences for cybersecurity education.

Global Expansion: Multi-language support for international users. Region-specific scam pattern recognition. Localized security advice and resources. Partnerships with consumer protection agencies worldwide.

Privacy & Security Enhancements: On-device processing option for sensitive content. End-to-end encryption for uploaded screenshots. Automatic PII redaction before analysis. Compliance with GDPR, CCPA, and other privacy regulations.

Integration Ecosystem: Slack/Teams integration for workplace protection. Email client plugins (Gmail, Outlook). SMS filtering integration for mobile devices. Integration with password managers and security suites.

Our vision is to make Guardian Nova the go-to global platform for scam protection, helping millions of people stay safe in an increasingly digital world.

Built With

Share this project:

Updates

Submission history