Artificial intelligence (AI) is moving fast and becoming more integrated into everyday life.
As these systems become more powerful, a critical challenge has emerged: understanding how they behave once they leave the experimental lab and are used by the general population.
Hayden Helm, Engr ‘18 (BS), ’18 (MS), is working to solve that problem.
As founder of Helivan, a San Francisco-based research and software company, Helm develops tools that help organizations study, monitor, and understand the behavior of AI systems as they interact with real users and real-world environments.
The work sits at the center of one of artificial intelligence’s most pressing questions: How can we trust systems that continuously adapt and evolve?
“One of our core ideas is that as language models or agents become more powerful and more personalized, understanding their behavior and when it changes is going to matter a lot,” Helm said.
The company represents the latest chapter in a career that began at Johns Hopkins, where Helm earned both his bachelor’s and master’s degrees in applied mathematics and statistics through the university’s dual-degree program.
“One of the things I appreciated the most about Hopkins was how easy it was to dive deeper into the topics that interested me,” Helm said. “The more I learned, the more interesting the questions became.”
After completing his master’s thesis under Carey Priebe, professor in the Department of Applied Mathematics and Statistics and member of the Data Science and AI Institute, Helm joined a DARPA-supported research project in the Department of Biomedical Engineering. Within a year, his work attracted the attention of Microsoft Research, where he spent the next three years studying machine learning systems and how they adapt to changing conditions. That research ultimately inspired Helivan.
“The real person using the system is never exactly like the training data,” Helm said. “So, how do you adapt as quickly as possible? That’s a very real problem.”
One of Helivan’s first products, AgentWatch, helps users monitor how AI systems behave over time and across different contexts. As AI tools become increasingly personalized and autonomous, Helm believes understanding and monitoring their behavior will be just as important as improving their performance.
The questions that drive his work today trace back to his time at Johns Hopkins.
Originally drawn to physics, Helm discovered statistics as an undergraduate and found himself captivated by the field’s blend of mathematical rigor and real-world uncertainty.
“I always had an inclination toward math,” he said. “Statistics was approachable, but it also opened up this really rich world. Once I got into it, I found myself wanting to keep exploring.”
Courses taught by John Wierman, professor emeritus, and Avanti Athreya, associate research professor of applied mathematics and statistics, helped shape that path.
More advanced coursework, including statistical theory and graduate-level classes with Priebe, provided the mathematical foundation he continues to rely on today.
“These courses gave me the tools to think clearly,” Helm said. “You start to see the world through probabilistic and statistical concepts.”
That perspective has proven especially valuable in machine learning, where data, users, and environments rarely behave as neatly as theory predicts. The ability to reason through uncertainty—a hallmark of applied mathematics and statistics—continues to inform both his research and his approach to building technology.
For current applied mathematics and statistics students interested in artificial intelligence, Helm’s advice is simple: start building.
“Get your hands dirty. Work on problems. Touch data. Touch software,” he said.
Looking back, Helm sees a direct line between the education he received at Hopkins and the work he leads today.
The mathematical thinking, curiosity, and problem-solving mindset he developed in the department continue to guide him as he explores one of the fastest-moving fields in technology and help shape how the next generation of AI systems are understood.