The students admit they are not the best dancers, but that did not matter to judges at the BigRed//Hacks last month, when they named the Carnegie Mellon University team first in two categories.

Athiya Deviyani, Abuzar Khan, Alex Li and Ranadeep Singh — all first-year students in the Master's in Artificial Intelligence and Innovation (MSAII) program — went into the Cornell University hackathon with a strategy of not taking it too seriously. But their dance app, Twerk Out, ended up a serious winner as the best health hack and the most creative use of Twilio.

"We worked well as a team and it all came together at the end," Singh said. "I would have never gone for such a 'nonpractical' idea myself, but it worked."

Among the oldest and biggest hackathons, Big Red//Hacks is open to university students across the world. This year more than 300 students participated, but in-person participation was only open to Cornell University students. Others, like the CMU team, participated virtually. During the hackathon, teams had 40 hours to solve a problem using technology. At this hackathon, teams were asked to come up with a way to encourage people to live healthier lifestyles while also reducing health insurance premiums.

The team took over the Language Technologies Institute lounge at 300 South Craig Street and settled in for the long haul. On the first night, they made little progress as they prioritized a homework assignment due the same day. They allowed themselves to sleep the first night and get some rest. But as the second day started, so did the serious work. The team hustled 24 hours straight.

"It was basically a nerdy sleepover," Deviyani said "There's pizza. There's food. There's caffeine. But there's no sleeping."

And there was lots of coding and dancing.

In Twerk Out, two players compete against each other to see who has the best dance moves. The players join the app via video and take their best shot at performing the dances suggested on the screen. Players score points depending on how successful they are at the dance move. The players can see each other dance and their scores, and an AI tracks how many calories they are burning.

As the team brainstormed ideas, they focused on fun. The app had to be fun to use, but for the team, it had to be fun to code. Twerk Out fit the bill nicely. The team used several AI tools to allow the app to track how players move, score them and determine the calories burned. About 10 open-sourced technologies worked together to make it possible.

"All those tools were things I encountered for the first time," Khan said. "And I found the best way to find the best thing that works is to go into the deep and hope you don't drown."

No one drowned making Twerk Out, but the team did start to flail a bit when it realized it didn't have the data it needed to train Twerk Out to recognize and score popular dance moves, like flossing. Some of the machine learning libraries available to the team weren't integrating with the app, so, with four hours to project submission — after not sleeping for nearly 24 hours straight — the team had to record their own dance moves as training data.

"We were terrible training models," Khan said about the team's dance moves.

Time was running out as the team raced to finish the app. The last 30 minutes were chaotic. With minutes on the clock, they hoped everything would work and recorded the demo video with only 15 minutes left.

"Near the end, we were really pressed for time. We just started throwing out ideas and doing anything we could to make it work," Li said. "It was hilarious."

And then it was done. The jury of companies and Cornell professors loved Twerk Out. It was a success, and the team relished the jury's validation of their hard work.

"I think a lot of people take hackathons too seriously," Deviyani said. "And I don't think people expected a bunch of grad students from Carnegie Mellon to create a dance app called Twerk Out."