Inspiration
AI has made UPI and banking scam messages look polished, urgent and real. Basic spam filters check single messages and send too many alerts, so users ignore them. We wanted a tool that spots the whole scam pattern, explains it simply, and warns only when it matters.
What it does
ScamShield AI analyses a suspicious message or payment request that the user submits.
- Extracts links, UPI IDs and urgency or impersonation cues
- Groups them into a likely scam workflow
- Gives a risk level with confidence: potentially fraudulent, suspicious, or no obvious indicators (not a guarantee of safety)
- Explains the evidence in plain English, with basic Tamil support
- Suggests safe actions and suppresses duplicate or weak alerts
How we built it
React dashboard, FastAPI backend, rule-based indicators plus a scikit-learn text classifier, similarity-based grouping, SQLite with minimal records, and synthetic scam examples.
Privacy and safety
It analyses only submitted evidence, never accesses bank or UPI accounts, masks identifiers, and never accuses on one weak indicator.
Challenges
Building realistic synthetic data, handling Tamil and mixed-language text, and cutting false alerts.
What we learned
Looking at the whole scam workflow works better than matching single keywords, and clear explanations build trust.
What's next
More Indian languages, voice and QR scams, and integrations only with proper permission.
Log in or sign up for Devpost to join the conversation.