ScamShield AI

Inspiration

Scam messages are everywhere — fake bank alerts, prize messages, job offers, delivery notifications, and suspicious links. As a beginner studying Artificial Intelligence, I wanted to build something that uses AI to solve a real-world problem while also helping people understand why a message may be dangerous.

That idea became ScamShield AI.

What it does

ScamShield AI is an AI-powered scam detection web application. Users can paste a suspicious SMS, email, WhatsApp message, job offer, or online offer and receive an easy-to-understand security analysis.

The application provides:

Risk Score: 0–100 Risk Level: Safe, Suspicious, or High Risk Scam Category Highlighted Red Flags Simple explanation of why the message may be suspicious Recommended actions to help the user stay safe

Instead of only saying "this is a scam," ScamShield AI focuses on explaining the reasoning behind the result.

How we built it

The frontend was built with Next.js, React, TypeScript, and Tailwind CSS.

When a user submits a message, the application sends it to a Next.js API route. The API sends the message to Google Gemini, which analyzes the content and returns a structured result.

I used Zod to validate the AI response before displaying it in the interface. This helps make sure the response follows the structure expected by the application.

The main flow is:

User Message → Next.js API → Google Gemini → Structured Analysis → Zod Validation → Results Dashboard

The application is deployed as a live web app using Vercel.

Challenges we ran into

One of the main challenges was making the AI response predictable enough for the application to use reliably. AI responses can sometimes contain unexpected values or missing information, so I used a structured Zod schema to validate the response before displaying it.

Another challenge was making the red flags useful instead of showing only a general warning. I designed the application to identify suspicious phrases from the original message and explain why each phrase may be a warning sign.

I also faced TypeScript and deployment issues while moving the project from local development to Vercel. Debugging these issues helped me understand the importance of production testing.

Accomplishments that we're proud of

I am proud that I turned an idea into a working AI-powered web application during the hackathon.

ScamShield AI can analyze real messages, assign a risk score, identify scam categories, highlight red flags, explain the reasoning, and provide recommended actions.

I am also proud that I successfully connected the AI model to a full-stack Next.js application, validated its responses with Zod, and deployed the project to a live production environment.

What we learned

This project taught me that building an AI application involves much more than simply connecting an AI model.

I learned how to work with AI APIs, structured responses, Zod validation, API routes, error handling, environment variables, GitHub, and Vercel deployment.

Most importantly, I learned how to take an idea, break it into smaller technical tasks, test each part, and turn it into a working project.

What's next for ScamShield AI

The next step for ScamShield AI is to make the analysis more comprehensive and useful.

Future improvements could include more scam categories, stronger scam-pattern detection, support for additional types of suspicious content, improved explanations, and more detailed security guidance.

I would also like to continue improving the user interface and make the tool even easier for people to understand and use.

Built With

• artificial
• css
• gemini
• google
• intelligence
• next.js
• react
• tailwind
• typescript
• zod