Inspiration
Phishing and scam messages are becoming increasingly convincing. Many users struggle to identify suspicious links, urgent requests, fake rewards, and messages that attempt to steal personal or financial information.
We wanted to build a simple security tool that helps everyday users understand why a message or link may be suspicious before they interact with it.
What it does
DefenShield AI is an explainable, privacy-first phishing and scam triage prototype.
Users can enter a suspicious message or URL. The application analyzes the content using local Python-based heuristics and identifies potential warning signs, including:
- Urgency and pressure tactics
- Credential and OTP requests
- Financial manipulation
- Suspicious URLs
- Social engineering patterns
- Potential impersonation indicators
The tool generates a risk score, displays detected findings, and provides actionable safety recommendations.
How we built it
The application was built using Python and Streamlit. Its backend uses a deterministic, rule-based analysis engine for text and URL inspection.
The processing is performed locally. The application does not open submitted URLs, send user messages to external services, or perform unauthorized scanning.
The project was developed as a defensive cybersecurity prototype focused on accessibility, explainability, and privacy.
Challenges we faced
One of the main challenges was designing useful risk indicators without making unsupported claims about whether a message is definitely malicious.
We therefore focused on explainable findings and recommendations instead of presenting the result as a guaranteed security verdict.
Another challenge was keeping the application simple enough for non-technical users while still providing useful technical information.
Accomplishments
- Built a functional phishing and scam triage application
- Implemented explainable risk scoring
- Added suspicious URL analysis
- Added safety recommendations
- Included local session history and report export
- Designed the application for defensive and educational use
What's next
Future improvements could include:
- Testing against a larger labelled dataset
- Adding calibrated machine-learning models
- Integrating trusted threat-intelligence sources
- Supporting additional languages
- Adding browser and email integrations
- Improving detection evaluation and reducing false positives
Important limitation
DefenShield AI is an MVP and should not be treated as a replacement for professional security tools or verified threat-intelligence services.
Log in or sign up for Devpost to join the conversation.