Inspiration
Honestly, the idea for Counterfeit Eye came from a video that we saw online, about a common scam in online markets such as Facebook Marketplace. These scammers show up intending to buy expensive items using fake banknotes, and since we do not have any accessible tools to check their authenticity, we have no way to find out.
We thought it would just suck if that happened to you, your hard-earned money gone because of one mistake. We wanted to change that. The inspiration was really about leveling the playing field and giving everyone a tool that’s simple, fast, and actually works to bring some trust back into using cash. We just wanted to build something genuinely useful that our friends and family could use, and protect our underrepresented community from these scams.
How we built it
For the most part, it was with Swift in Xcode, since it is an IOS application, we have to design and develop in Swift, a language barely any one of us knew. We struggled for a few hours but we made it.
Python was our best friend. Using Torch library we trained our AI model, and we also used timm library for a pre-trained model that would be more efficient to train in a limited amount of time. We also used Flask to make an API for our AI model, and make it usable from the mobile application.
We added a feature which was locating the nearest police department after detecting a note was counterfeit, we used Gemini's API where we send a prompt containing the current coordinates, and the AI sends back the nearest police department in JSON Format.
The Challenges
Coding in Swift was definitely a challenge, because we never used this coding language and it was the most part of our project. We ran into a lot of errors, but we made it out alive.
Another big challenge was that there was no dataset that satisfies our needs to make this AI, so we had to build one from scratch. I took a picture of the 100 dollar bill in every possible lightning condition, until the dataset was well over 1200 in total.
The last challenge was that I had a $15 MacBook, that means for training it's a nightmare. We mostly used an under trained model for testing, but the final model had to be strong and accurate. Luckily, our model finished training in the last 5 minutes before the deadline.
Log in or sign up for Devpost to join the conversation.