AI Scam & Phishing Detection System

Inspiration

Online scams and phishing attempts through SMS, emails, WhatsApp messages, fake links, and unknown phone numbers are increasing rapidly. Many users fall for these scams not because of lack of intelligence, but because scam messages are designed to create urgency, fear, or false trust.

Most existing solutions act like black boxes — they label something as “scam” without explaining the reason. This project was inspired by the need for a transparent and explainable system that helps users understand why something is risky, not just that it is risky.

What it does

AI Scam & Phishing Detection System analyzes user-provided inputs such as:

Text messages (SMS, WhatsApp, email content)
Website links and URLs
Phone numbers

The system evaluates the input using rule-based scam detection logic and assigns a risk score and risk level. It then generates a clear, human-readable explanation of the detected red flags so users can make informed decisions.

The AI does not decide whether something is a scam — it only explains the results produced by deterministic rules.

How we built it

The project is built with a simple, efficient, and privacy-focused architecture:

A Flask backend handles all analysis requests
Rule-based logic detects scam patterns such as urgency cues, OTP or KYC requests, suspicious domains, URL shorteners, and phone number anomalies
A scoring system converts detected red flags into a risk score
An AI explanation layer (Gemini) converts technical findings into easy-to-understand explanations
A lightweight frontend allows users to quickly test messages, links, and numbers

No user data is stored; all analysis happens in real time.

Challenges we ran into

One major challenge was avoiding over-reliance on AI. Using AI directly for classification would reduce transparency and increase cost. Designing a system where AI only explains decisions required careful separation of responsibilities.

Another challenge was creating detection rules that are conservative, explainable, and still useful across different scam formats and languages.

Accomplishments that we're proud of

Built a fully explainable scam detection system
Kept AI usage minimal and cost-efficient
Designed a judge-friendly system that is easy to demo and understand
Ensured user privacy by avoiding data storage
Delivered a complete, working solution within hackathon constraints

What we learned

This project helped us understand how to build explainable AI systems, design rule-based security logic, and balance automation with human trust. We also learned how to structure systems that prioritize clarity, cost control, and real-world usability over hype.