our web ui

Inspiration

In 2025 80 billion dollars from scams that started with fraudulent SMS Messages. SMS fraudulent messages continue to become more advanced every year. We were inspired to build this web-site in order to help reduce the amount of financial damage that comes from fraudulent activity on the internet over the following years.

What it does

It detects fraudulent sms messages which a user can upload into the website’s textbox and click the green button. The user will get a score of how accurate the message is fraudulent or not. If you get a score of below 50%, then the message isn’t fraudulent. If the score you get is above 50%, it is fraudulent. This website does have another 2 button boxes that could create fraudulent and non-fraudulent examples so it will help you guys compare and understand how your real life example of messages might be fraudulent or not.

How we built it

We built our AI agent to detect fraudulent and non-fraudulent messages by inserting dataset message examples for the AI to be trained to follow a pattern of how a real life fraudulent message example might be looking like. We created another coding set to confirm whether the real life example is almost accurately a fraudulent message or not. Our accuracy results averaged around 86%. We trained our AI agent not only by a trained data but we also made a program to have another AI throw sample fraudulent texts so that it can detect examples so that it will work in the modern world. We also had assistance with github's copilot agent.

Challenges we ran into

We faced many challenges such as our accuracy because our accuracy could only get raised to 86%. If we had more time we could've trained it more to have a better accuracy. We also had to fix many errors by telling the copilot agent what to do when it ran to a block.

Accomplishments that we're proud of

We were able to make an AI model that accurately detects if weather an SMS message is fraudulent or not.

What we learned

We were able to create an AI model which could be able to learn from mistakes it made and correct itself, and using data sets that we found online we were able to train the AI model to accurately detect whether an SMS message was fraud or not. We also implemented an AI training simulation where we ran another AI that would give prompts to our AI Model and would test and teach the AI model to detect fraudulent messages more accurately.

What's next for ScamSheild

Another idea that could be added in the future to Scam Shield would be the ability for users to also input emails and other forms of messages to our website; which our AI model would be able to detect whether or not these emails and messages from other apps are fraudulent or real.