Nucleo calls itself the first agentic platform for oncology. What its founders mean by the word is an agent that coordinates specialized imaging models within a defined workflow, while clinicians retain the final say.
Founded in 2025 by mathematicians Angelica Iacovelli and Luca Pegolotti, both trained at Politecnico di Milano, the company analyzes CT and MRI scans to extract body-composition metrics and handle tumor detection, segmentation, and classification. It began with lung nodules and is broadening from there. Nucleo says its automated analysis runs 2,500 times faster than manual segmentation.
The pair's route into the problem ran through conversations with clinicians at Stanford, where Iacovelli was modeling cardiovascular blood flow and Pegolotti was researching at the School of Medicine. "We were talking with a lot of doctors to understand their workflows," Iacovelli says, "and we realized that there was a big problem in radiology."
Nucleo’s co-founders talked with The Infinite Loop about how they found their niche and what’s on the horizon.
TIL: You call Nucleo agentic. To what extent are your agents autonomous?
Pegolotti: The reality is that the clinician always has the final word. We provide an indication of what the image looks like, what data we can extract from the image, but in the end it’s the clinician who decides what to do with the data that we provide.
Iacovelli: Being agentic doesn’t mean that we decide everything. We automate a bunch of tasks. At the same time, the final report is reviewed by the doctor, who decides whether they agree with our findings.
TIL: So what is the agent actually doing?
Pegolotti: The AI models that we trained ourselves, or that we make available, are exposed to an agent. We have some workflows that require different AI models to interact with each other. Some of the models are more classic, like computer vision-based. Some are a little bit more cutting-edge, let's say. The latest ones, for example, are based on vision transformers for segmentation. But the agent handles the orchestration. We have different models, and the agent decides which one to call based on the workflow.
TIL: What sets Nucleo apart from existing clinical imaging tools?
Iacovelli: We started with a specific wedge product that enables us to provide a total body composition assessment of the patient, an approach no one has taken so far. In addition to the data points for a specific tumor, it also provides a complete nutritional and metabolic picture of the patient. We’re also building a framework that enables us to generalize across different tumors. We don’t want to build classic computer vision models that can detect a single tumor, a point solution. We’re building models that can generalize across a lot of different diseases.
TIL: You describe a framework for different diseases. How far beyond cancer does it go?
Iacovelli: It’s already expanding. We work with oncologists, radiologists and surgical oncologists, but we also work in obesity and weight loss, where body composition assessment is really important, as well as in geriatrics.
"A really big barrier to entry"
TIL: What has been hardest since you started Nucleo?
Iacovelli: Something that all healthcare companies struggle with is how slow everything is. Startups, by definition, need to be really fast to be compatible with the timelines of VC funds. At the same time, we operate in a space that is really slow. Sales cycles can take years, and it’s a heavily regulated space. You really need to have the resources and the time, and try to be as fast as possible. There’s a really big barrier to entry in this space.
TIL: How did you get from a first conversation with a hospital to a deployment?
Iacovelli: What has enabled us to talk with all these hospitals has been having the first few champions, doctors who have believed in us. The next most effective thing is doctors talking to other doctors. Warm intros or word of mouth work way better than doing cold outreach to a hospital or an institution you’ve never worked with.
“Not something I would make a movie of”
TIL: We run a newsletter format called The First Customer, where we ask founders about the moment a company got its first paying customer. So, plainly: who was yours, and how did the deal happen?
Iacovelli: We can’t disclose the names of the hospitals we work with, but we can tell the story. The first customer we onboarded, the doctor we met, [came] through an introduction from an investor. A lot of doctors we have met have been through investors. This doctor was really excited about the technology. We didn’t have much at that point. But he was super excited, and initially we had this relationship where we were building [the product] following his recommendation. Every startup initially needs to have a design partner, and they were our design partners. They still are. By creating that relationship with a hospital, at that point they will trust you. They will know that you're fast, that once they have a problem, you're always there to answer. And once that trust is built, it's way easier to sell.
TIL: What convinced them?
Pegolotti: I think they saw how we operate. What really makes a difference at this stage is that you have to show that you’re quick at iterating. If they have a specific request, you come back a week later having implemented that feature, and that really makes a difference. We have a background in mathematics. So sometimes it’s a little bit hard for us to understand, does this even make sense from the clinical perspective? Having these interactions with the doctors is very valuable.
Iacovelli: You need to imagine doctors working with these really old tools and interacting with big vendors that are really slow. So when they work with fast-paced companies, they’re like, “wow, this is fantastic.” It's really important to be fast and to show that if they have a problem, we're really quick, and we can build something incredible in a few days. Which is something they're really not used to, working with big players.
TIL: Was the close itself spectacular?
Iacovelli: It was not spectacular. It was not something I would make a movie of.
Pegolotti: It was almost all virtual. We’re often in the US and Switzerland, and it’s difficult sometimes to just meet in person.
Iacovelli: We always try to go in person for training or to create a relationship, but in this case the contract was closed virtually.
TIL: You won first place in the Medical Imaging category at the Nebius AI Discovery Awards. What’s your plan with the credits?
Pegolotti: We believe the future of radiology will not be built from hundreds of disconnected AI models, but from unified foundation models capable of reasoning across imaging, clinical history, and multimodal patient data. We will use the Nebius compute to lay the groundwork for these models, starting with oncology foundation models for pan-cancer detection, lesion quantification, and classification.
Disclosure
Iacovelli and Pegolotti spoke with The Infinite Loop at the Nebius AI Discovery Awards ceremony in London, where Nucleo took first place in the Medical Imaging category. The Infinite Loop is a Nebius editorial project. Editorial decisions are made independently.
