Cole Roberts, Sc.M. ’16 combines AI and medicine to empower physicians

How did Tulip come about?

 Roberts: Before Tulip, I worked as the first product hire at two early-stage companies. One was Unlearn.ai, which uses generative modeling for clinical trial control arms, particularly in Alzheimer’s disease.The second company, Atropos Health, focused on using real-world data from electronic medical records for bedside care and drug development. It was inspired by Stanford’s Green Button Project. The idea was to empower physicians with real-time insights by analyzing decades of EMR data. For instance, a doctor could ask how similar patients were treated and get actionable insights within hours instead of months.

Both experiences taught me the ins and outs of early-stage startups and product development. With Tulip, I wanted to build something impactful for doctors. Over the past year, we’ve been working on an AI-powered tool called Dr. Tulp—an AI “doctor for doctors.”

What exactly does Dr. Tulp do?

It’s designed to address the overwhelming amount of medical knowledge that doubles every 75 days. Doctors often struggle to keep up with the latest research and connect it to patient care. Dr. Tulp acts as a specialized assistant, providing summaries of patient data to help doctors get up to speed quickly. It also offers insights into medical topics and treatment options, differential diagnosis trees to help identify potential conditions based on patient criteria as well as integration with EHRs to streamline responses to patient inquiries. We see it offering significant ROI for both physicians at the bedside as well as system-wide administrators.

We’ve trained the model on the entire corpus of medical knowledge, making it a peer-level assistant for doctors. It even outperforms the average doctor on licensing exams, scoring in the high 80^th or low 90^th percentile. We’re planning a wider launch in January.

The name, Dr. Tulp, has an interesting origin.

Yes, it’s inspired by Rembrandt’s painting The Anatomy Lesson of Dr. Tulp. The painting depicts a public anatomy lesson led by Dr. Nicolaes Tulp, who was a renowned physician and the head of Amsterdam’s Guild of Surgeons in the 1630s. Much like the painting’s theme of sharing expertise, the AI Dr. Tulp aims to be an expert assistant that supports and educates doctors in real-time. The product’s name embodies our mission of bringing cutting-edge expertise to doctors, much like how Dr. Nicolaes Tulp shared knowledge centuries ago.

What do you see as the biggest challenges facing public health today?

I’ve been thinking about this a bit, and I feel like the pandemic really eroded a lot of trust. And what made the pandemic so hard was that it exposed a huge gap between how health care workers approached the problem—acknowledging its complexity—and how the public perceived it. From a public standpoint, it felt like: Here’s this massive, urgent problem; why aren’t you fixing it already? And the truth is, we developed a vaccine in about 12 months. That was remarkable! But where we really struggled—and still struggle—is communication.

We need to get so much better at communicating effectively, and I think leaders like Dean Jha at Brown, who’s great at this, are helping move us in the right direction. Still, rebuilding trust is going to take a lot of work over the next decade or two. I just hope we don’t have another pandemic-level crisis anytime soon, because I worry we’d be in an even worse position now to handle it.

During my time at Brown, I took a course with Chris Koller, the former Rhode Island Insurance Commissioner. It was an eye-opener about how insurers, providers and patients interact. If more people understood how the system works, it could help demystify many frustrations and improve navigation.