Duke AI Hackathon Showcases Student Ingenuity in a Fast-Evolving Field

It was just after 7:30 pm on a Saturday evening, and Sakshee Patil felt like she had hit a brick wall. She’d been trying to make her team’s idea for a new AI-powered product work for hours, but every iteration managed to find a new way to fail.

It didn’t help that her mind was nearing a state of numbness. The 2024 Duke AI Hackathon was already 26 hours into its two-day-long schedule.

But then something magical happened.

“I got a single simple query to work end-to-end, and that was the breakthrough we needed,” said Patil, a first-year student in Duke’s Master of Engineering in Artificial Intelligence program. Even though she’d only slept for four hours that day, the rush from seeing it work after countless attempts was enough to combat the weariness. “That success fueled me to push through the night, and by 3:46 am, I had a full version ready to present. The timing was tight, but that moment of seeing it all come together was unforgettable.”

Part of her team’s challenge was that the project they were pursuing relied on a feature released by OpenAI (the makers of ChatGPT) only a couple of weeks before. Called “Swarm,” the software quickly and autonomously builds a group of AI agents, each of which is responsible for a different task, to tackle multifaceted requests. Their idea was to add a natural language processor on top of Swarm so that anybody could easily build and test ideas that work with any program’s existing application programming interfaces (APIs).

For example, they demonstrated swarms built by their software that:

• Fetch the latest Bitcoin price, 24-hour price change and recent news for analysis
• Retrieve the top 20 songs on Spotify based on user-specified genres and provide a playlist summary with audience insights
• Check if a given Twitch streamer is currently live with an option to send a humorous message if they are offline
• Search X for recent discussions on a topic and summarize the main viewpoints
• Generate promotional content for a YouTube video, including a catchy title and description based on the video transcript

“I was impressed by the quality of the ideas presented, and even more importantly, the completeness of their ‘minimum viable product’ given such a limited timeframe to put the idea together,” said Rubin Dong, founder and CEO of Scale Social AI who graduated from Duke Engineering in 2012. “The projects were truly inspiring, showcasing the remarkable talent and creativity of Duke students as they tackled real-world challenges.”

The hackathon included many projects including “Inception Swarm,” described above, which won the overall grand prize for the event. In the event’s second year, it attracted nearly 150 students, who banded together into over 30 teams to bring AI-inspired ideas to reality.

The program kicked off at 5:30 pm on a Friday with training sessions that educated participants on a variety of AI-related topics. By around 9:00 pm, groups were formed, ideas were knocked around and the hacking had begun. Throughout the next two days, teams worked together—and with experts and mentors from industry and Duke’s faculty—to build software that could make a difference in one of five tracks: education; health and wellness; productivity and enterprise tools; art, media and design; and social impact and climate.

“Compared to last year’s hackathon, the level of technology fidelity and novelty was significantly higher,” said Jon Reifschneider, director of the Master of Engineering in Artificial Intelligence program. “That tracks with how the field has evolved and the greater availability and performance of AI tools.”

To help orient participants to the latest in AI tools, technical sessions and a technical help desk were available to everyone.

“Hackathons are an immense opportunity to level-up on your knowledge and skills,” said Brinnae Bent, a faculty member in the Master of Engineering in Artificial Intelligence program. “Technical sessions on LLMs, natural language processing, computer vision and raspberry pi controllers led by Duke student organizers and faculty kicked off a weekend of learning and skill building.”

By 9:00 am on Sunday—just 36 hours after the kickoff—teams presented their working products to a panel of 20 judges, almost all of whom came from industry or practice. Each track had one winner, and the five track winners then competed for the grand prizes.

“The hackathon wasn’t just about winning, though. It was about pushing through challenges and learning as much as possible along the way,” said Patil.

Vivek Rao, director of the Master of Engineering in Design & Technology Innovation program, agreed and added, “Every team walked away with an incredible prototype and some rich, nuanced feedback from our industry-leading judges.”

While “Inception Swarm” took home the grand prize, $1500 in credits to OpenAI and a guaranteed investor feedback meeting with 2ndF, there were other incredible ideas that took form during the hackathon as well. One of them, called “Flashback,” sounds like something straight out of an episode of the Netflix series “Black Mirror.”

What began as an idea for people to locate items like their keys evolved into an app that can help people remember just about anything from their recent past. By recording footage from the user’s point-of-view and sending it to the cloud, the program helps recall events through a web interface that allows the user to ask questions about who they met or important information about people they talked to.

But why stop there? One of the judges pointed out that such a system could be transformative for people with short-term memory challenges, helping them maintain their independence and quality of life.

“This feedback expanded our perspective, showing us how our educational tool could evolve to potentially help countless individuals with memory-related conditions,” said Michael Dankwah Agyeman-Prempeh, a student in Duke’s Master of Engineering in Design Technology and Innovation program. “These moments of discovery reminded us why hackathons are so exciting—you never know when a simple idea might blossom into something that could genuinely change lives!”

“The Duke AI Hackathon has been a truly transformative experience for me,” added Agyeman-Prempeh. “I’ve learned invaluable lessons about collaborative problem-solving, rapid prototyping and the power of interdisciplinary teamwork.”