Machine Learning Best Project, DeerHack 2024

Nymna Alert minimizes response time with real-time monitoring and an instant alert system through a mobile app. Instead of relying on someone watching a camera feed, it detects anomalies automatically and pushes an alert straight to a phone, cutting the time between something happening and someone finding out.

We built it in 48 hours at DeerHack 2024, Deerwalk's annual hackathon, as a 5-person team, and it went on to win Machine Learning Best Project.

Since this was our first hackathon, the main challenge was dividing tasks among team members, and figuring out when each of us would actually do the delegated work, which ultimately kept us up for the whole two days.

The bigger challenge was technical: training the computer vision model with the resources and data we had. Accuracy was hard to hold up given how little annotated data we had access to, we underestimated how much we'd need, and a whole night of work was lost because the data wasn't properly annotated. Annotation itself also ate up a lot of time.

We'd originally scoped a broader anomaly detector covering fights, violence, and weapon detection, but the time frame only let us ship a narrower version. Despite all that, it was an incredible, thrilling experience.