ScamSense

Inspiration

I've seen why that's such a big problem because I've almost fallen for scams myself. I've gotten random texts saying my car would be towed, that I owed money, or that I needed to pay a fee immediately. One of them looked pretty convincing, and for a second I actually stopped and wondered if it was real. I'm only 18 and spend a lot of time online, so if I had to think twice about it, imagine how stressful something like that can be for someone much older.

Older adults are often targeted because scammers know they may be less familiar with newer types of fraud and may not have someone nearby to ask. People lose billions of dollars to scams every year, and a lot of that happens simply because there isn't an easy way to check whether something is legitimate in the moment. Most people end up searching online and hoping they find the right answer.

That's why I built ScamSense. Instead of guessing or searching through websites, someone can paste in a suspicious message and get an instant answer, understand why it was flagged, and know exactly what to do next.

What it does

ScamSense helps people quickly figure out whether a text message, email, or phone call transcript is a scam.

Users can paste in any suspicious message and within a few seconds get:

A verdict: Scam, Likely Scam, Suspicious, or Safe A confidence score The type of scam it looks most similar to Specific red flags found in the message Suggestions on what to do next, such as blocking the sender, reporting it, or contacting the actual organization

ScamSense also includes a community feed where users can anonymously share scams they've received so others can stay informed about new tactics scammers are using.

How we built it

We built ScamSense using Next.js 14 and TypeScript for the frontend and backend.

For scam detection, we use the Groq API with a large language model to analyze messages and return structured results including a verdict, confidence score, scam category, red flags, and recommended actions.

Anonymous community reports are stored in Supabase and displayed in a live feed. To protect privacy, only a shortened version of the submitted message is saved.

Tailwind CSS and Framer Motion were used to create a simple interface with smooth animations and color coded verdict cards. The project is deployed on Vercel.

Challenges we ran into

One challenge was getting the AI model to consistently return data in the exact format we expected. Sometimes it would include extra explanations or slightly change the structure of its response, which caused issues when parsing the results.

To solve this, we improved our prompt and added validation checks so the application could still handle unexpected outputs gracefully.

Another challenge was designing the interface. ScamSense is meant for older adults, so we wanted it to feel easy to use and reassuring instead of overwhelming or overly technical.

Accomplishments that we're proud of

Built a working application in less than a week Successfully identified common scams such as IRS impersonation, Medicare scams, phishing attempts, lottery scams, and romance scams Created an interface simple enough for someone with little technical experience to use Added a community reporting feature so users can help warn others about emerging scams

What we learned

While building ScamSense, I learned that scam prevention isn't only about improving detection accuracy but also about making help available at the exact moment someone is unsure what to do. We also learned how effective modern language models can be for classification tasks even without custom training. With careful prompting and output validation, they can provide surprisingly useful results for real world problems.

What's next for ScamSense

Browser extension that warns users about suspicious emails Support for analyzing phone call transcripts SMS forwarding so users can text suspicious messages to ScamSense and receive an instant response Family accounts that allow relatives to monitor scams flagged by older family members Support for multiple languages Partnerships with senior centers and fraud awareness organizations

Built With

• framer-motion
• google-gemini-api
• groqapi
• next.js
• supabase
• tailwind-css
• typescript
• vercel