Since they returned to Madison in mid-January for the spring 2026 semester, roommates Ruffin Bryant and Noah Kalthoff have settled into a familiar—if busy—rhythm.
Mornings: Class, like most of their fellow biomedical engineering majors at the University of Wisconsin-Madison.
Afternoons and evenings: Working full-time hours on OpScribe-AI, a startup venture they’re hoping will turn into a viable career option by the time they both graduate in May 2027.
As the name suggests, OpScribe-AI is an artificial intelligence-powered tool for automated surgical video analysis and pricing. It’s designed to improve the accuracy and efficiency of surgical documentation and medical coding that are currently manually produced after every operation and used for pricing, insurance reimbursement and tracking patient health, while alleviating some of the administrative burden on surgeons. The latter is frequently cited among the leading sources of burnout in the profession.
“We literally met with a surgeon who said, ‘The worst part of my job is writing reports.’ It’s not why they went to med school. It’s not fun,” says Bryant, a junior from Greenwich, Connecticut. “We make it a one- to two-minute review.”
The nascent startup is an outgrowth of a project Bryant, Kalthoff and four other biomedical engineering students undertook during the fall 2025 semester as part of the Department of Biomedical Engineering’s undergraduate design program. The two had planned to work together on a project leading up to the start of the semester, when BME students form groups and rank their preferences from the list of projects sponsored by industry partners, academic researchers and community members.
AI was a shared interest. Kalthoff spent summer 2025 in Barcelona interning at an early-stage tech startup, gaining coding experience and helping to implement AI tools. Bryant, meanwhile, had played around with building a machine learning algorithm in spring 2025. Still, when the roommates got their first-choice project, proposed by Tuo Peter Li, a UW-Madison assistant professor of orthopedics, they had a moment of trepidation.
“We basically looked at each other and were like, ‘How are we going to do this?’” recalls Bryant.
Rather than training a new deep learning model from scratch, the students—on the advice of Biomedical Engineering Assistant Professor Dhananjay Bhaskar, their project advisor—opted for an agentic, AI-based approach, integrating pre-trained large language models and deep neural networks into a family of AI agents that collaboratively process surgical videos to generate standardized operative reports and standardized medical codes.
To avoid erroneous results—hallucinations, in AI industry speak—any information the AI agents aren’t certain of is left blank for the surgeon to complete. OpScribe-AI also pulls in information from the National Institutes of Health PubMed database to add context from medical literature and tailors the output to individual surgeons based on previous reports.
“That’s where the technical sophistication lies: building an agentic AI system that is able to reach out into literature, retrieve medical context from surgical reports and other medical publications, and leverage all that context to build an operative report that accurately summarizes the medical procedures,” says Bhaskar, who develops deep learning methods for biomedical discovery and joined UW-Madison in fall 2025 as part of the university’s RISE-AI initiative.
Importantly, the large language models in OpScribe-AI are open-source, self-hosted components that can work within a hospital system’s digital infrastructure while complying with HIPAA regulations.
In addition to saving surgeons’ time and mental energy, OpScribe-AI ensures more accurate billing; missed details cost hospitals when it comes to reimbursement.
“The adage is that if you didn’t document it, then it didn’t happen,” says Li, who along with Bhaskar is part of OpScribe-AI’s scientific advisory board.
To further develop their prototype, though, Bryant and Kalthoff need more robust data than the open-source videos they’ve used thus far. With that in mind, fellow biomedical engineering design project team members Lucy Wyse and Kendall Witt have pulled together information for the group’s Institutional Review Board application to access real surgical data that’s necessary to validate and improve the platform.
The design project team, which also includes Wylie Lu and Evan Matthews, is also working on an academic journal paper based on data collected from the UW Health Clinical Simulation Program, which stages training operations on cadavers and phantoms. By sharing their work and the simulated surgical data, the students hope to encourage more development at the intersection of AI and the biomedical sciences.
Since turning their design project into a business venture in mid-December, Bryant and Kalthoff have filled their calendars with pitches to potential investors and applications for startup accelerator and incubator programs. In early March 2026, they were accepted into the gBETA AI Startup Accelerator through Waukesha County Technical College’s Applied AI Lab.
The students plan to recruit software engineering expertise to their team, to improve the user interface, and to explore methods for leveraging data to uncover actionable insights for healthcare decisionmakers—trends during particular procedures, for example—that could amount to another revenue stream down the road.
“Who knows where this is going to take us, but we both agree that fail or success or somewhere in the middle, the lessons and the amount of stuff we’ve learned is going to be so applicable to the next steps in our career,” says Kalthoff, a junior from Delano, Minnesota. “It’s kind of a win-win, no matter what happens.”
Top photo caption: Ruffin Bryant, left, and Noah Kalthoff first met as freshmen during move-in day at Dejope Hall. Photo: Joel Hallberg