UTMB AI innovators win international hackathon with radiology viewer and teaching tool

What began as a vibe coding side project for a small team of developers at The University of Texas Medical Branch (UTMB) became one of 50 winners out of 4,096 entries in the Google DeepMind Kaggle hackathon. Vibe coding builds software through plain language prompts rather than traditional code - the user describes what they want, and AI does the rest.

It started with a question.

While attending an AI conference on campus late last year, James Weatherhead, an MD-PhD student, learned of the hackathon during a presentation and asked Dr. Peter McCaffrey, chief digital and AI officer at UTMB, for ideas about what they might create with Gemini 3, the required AI application.

McCaffrey said, "I wonder if it can build a radiology DICOM viewer."

James Weatherhead thought it was ambitious but worth trying. Digital Imaging and Communications in Medicine (DICOM) is the global standard file format for radiology. Viewers allow users to interface with MRIs, X-rays, CT scans, and other images.

What is vibe coding?

Vibe coding is a software development method that has users provide a description of what they want to create, then guide the "vibe" of the product by providing feedback in plain language, which leads to the next iteration of the product.

"After maybe three or four messages, a really bad-looking DICOM viewer was starting to form and we were like, whoa. This is unbelievable," James Weatherhead said. "And then over the next couple of weeks it just became a bigger and bigger thing."

That was the beginning of VibeRad, the winning project. With the added expertise of George Golovko, PhD, assistant professor of pharmacology and toxicology, and Jake Weatherhead, a senior software engineer outside UTMB who contributed on his own time, James Weatherhead pushed the project even further, transforming it from a viewer to a teaching tool.

From high school to residency, one tool

"I thought, 'I wonder if the AI tutor can actually teach me what I'm looking at?'" he said. "It was quite crazy. It actually did. But I'm a med student, and not everybody is a med student."

James Weatherhead then requested toggles for teaching levels: high school, undergraduate, medical student, and resident.

"The competition version of VibeRad was an AI teaching assistant for medical imaging," he said. "Students browse MRI series, annotate, measure, and ask the AI to explain what they're seeing. Lessons adapt from high school through residency. Simple. User-driven. One question at a time."

What's next for VibeRad

Even though the hackathon is over, James Weatherhead is already thinking about what comes next.

"I would love to see it released in a form where researchers can build on it and publish on vision-language model education, working with UTMB on how that gets done," he said. "Right now, a lot of imaging learning still happens through slides, PDFs, or board prep, because the clinical viewers themselves were built for practice, not for teaching. We can't zoom in on the actual images or revisit questions on our own time the way we can with an interactive tool."

He hopes to see more conversation and research around such teaching models, which could help fill learning gaps in medical schools and open new possibilities for medical education.

As much as the VibeRad team learned from the hackathon, winning was a bonus.

"When the submission deadline passed, I could see everyone's submission," he said. "So I was going through it for a couple of days and there were some really innovative ideas. I didn't really expect to win!"

The whole experience has given James Weatherhead a deeper appreciation for what the future may hold.

"What excites me most about AI is that it changes the distance between imagination and experiment," he said. "An idea that once would have required a team, a budget, and months of development can now become a rough educational prototype in an afternoon. VibeRad is a teaching prototype, not a clinical tool. But it shows something important. Clinicians, trainees, and other frontline staff can now take part in shaping solutions much earlier. That may ultimately change not only what gets built in medicine, but who gets to build it."

VibeRad is a DICOM viewer and teaching tool that allows users at various skill levels to learn from and about medical images.