Boston Scientific bought his startup for $290M — now he's building AI that admits it doesn't know
What's the deal? Boston Scientific acquired Cortex, a medical decision-making startup founded by scientist Peter Ruppersberg, for up to $289 million. The deal, announced in January 2025, comprised $239 million in cash plus up to $50 million tied to clinical milestones, per filings with the US Securities and Exchange Commission.
What did Cortex do? The startup spent roughly a decade turning complex clinical data into decisions doctors could use mid-procedure. Its main product, OptiMap, maps the heart's electrical activity to guide treatment of atrial fibrillation.
What's next? Rather than exit, Ruppersberg founded AmmonixDealroom has a profile for this one. Try Dealroom →, an AI startup building small, specialised models. Based in Blonay, Switzerland, it spent over a year in stealth before launching publicly on September 22.
Why the pivot? Ruppersberg's decade working with arrhythmias taught him a wrong answer stated confidently can be worse than none. "In medicine, you learn repeatedly that being right most of the time isn't enough," he told EXAMEDealroom has a profile for this one. Try Dealroom → (translated from Portuguese).
The product: Ammonix bets on the opposite of the industry's race toward ever-larger models. Its systems are small, trained on relatively little data, run locally, and can explain their decisions. "A fluent answer isn't necessarily a good decision," Ruppersberg said.
The architecture splits decision-making into two parts, mirroring how a cardiologist reads an electrocardiogram: a fast, specialised system decides, and a language model explains, asks, and converses. Crucially, it is designed to flag when it lacks enough evidence and pass the case to deeper reasoning or a human expert.
The founder: Ruppersberg began his career on a Nobel-winning research team, became a tenured professor and department head, and published in Nature and Science . Before Cortex, he founded Mentasys, an early AI product-search company sold to LycosDealroom has a profile for this one. Try Dealroom → in 2006.
The signal: While the AI industry measures models in trillions of parameters, Ruppersberg argues a few hundred carefully described cases can hold more useful information for a specific decision than millions of loosely related examples. His bet is that the next edge lies not in scale, but in models that know their own limits.
Read more: exame.com
Image credit: Generated with Gemini
