Raydar

Drawing flashlight on white board to test ML image recognition
Josh being in this position for 8 hours straight as he completed the text extraction model
9 hours into the night, everyone's tired. But still moving forward.

Inspiration

We were inspired by the cumbersome and inefficient ways in order to report lost or found items in the UIC's directory (https://tie.uic.edu/help-articles/lost-and-found-information/). Drawing from our own experiences of losing various items and never finding it, and the fact that Americans lose an average of 5 items every day, we decided to finally solve this issue.

What it does

Raydar is a dual machine learning application that analyses and matches uploaded sketches of lost items with found items. These found items are reported to the database by janitors, students, and anyone with a stable internet connection.

How we built it

We built Raydar using TensorFlow CNN for training the the first, image identifying model. For the second, text extracting model, we paired with scikit-learn and a text analysis method for extraction. We used NextJS in the frontend paired with ReactJS and ShadCN components with TailwindCSS for styling purposes. For authentication, we decided to use Clerk and our choice of database was Supabase. Auxiliary libraries used in training were spaCy and sci-kit learn for NLP, Fuzzy for removing typos, Levenschentein for semantic similarity.

Challenges we ran into

The biggest challenge for us was adapting to the strict timeline of 24 hours, when our team was accustomed to 48 hours of development time. Through the sleepless night, stuck in a room with our chargers and computers, it took real heart to break through the fatigue that wore down each of the team members.

Accomplishments that we're proud of

We’re proud to have met the deadline with a robust, optimized application with two effective models along with standout features such as a maps system, real time alerts, and a feed that unites owners with their items.

What we learned

We learned a lot about how to integrate various systems into a harmonious functioning application. We learned how to separate different application layers (front end, backend, data processing), we were able to learn how to scale and maintain a software with said modular approach. Effective communication and collaboration was a key skill that we needed to adopt through this hackathon and mastering them brought us closer as a team and growing life long skills.

What's next for Raydar

We want to apply Raydar to its true purpose, a lost and found application for UIC that is ahead of its time and makes losing items a little bit less stressful for the students.