He spent years teaching machines to read the immune system. Then he turned the same toolkit on something harder to map: the week between one therapy session and the next.
Luis Voloch is co-founder and CEO of Jimini Health, a New York company built on a simple complaint with the way therapy works: you see someone for fifty minutes, once a week, and then you are on your own until next Tuesday. Six days of silence. Jimini's answer is Sage - an AI assistant that lives in those gaps, reinforcing the exercises a licensed clinician assigns, never standing in for the human.
The distinction matters to Voloch, and he repeats it like a man who has watched the alternative go wrong. Sage does not diagnose. It does not freelance. It works under therapist supervision, nudging patients through cognitive behavioral techniques the clinician has already prescribed. The model is human-led, machine-extended - and Voloch insists that is the entire point, not a hedge.
He likes to describe the ambition in borrowed grandeur: a "Genentech-like company in mental health, powered by AI." It is a big claim from a young company. But Voloch has done the thing before where a strange technical idea becomes a billion-dollar enterprise, so he has earned a little swagger.
The most powerful AIs are not standalone products. They collaborate with human clinicians.
Before therapy, there was the immune system. In 2018 Voloch co-founded Immunai and became its CTO, applying machine learning to immunology and drug discovery - building, in the company's own framing, a vast computational map of how immune cells behave. The work was cited by MIT News as a model for how AI could accelerate drug development. The company climbed onto Crunchbase's unicorn board.
He ran the technology, sat on the board, and chaired the scientific advisory group until the end of 2022. Then he did the unusual thing. He left the unicorn he helped build and started again from zero, in a field with different rules and far softer data: the human mind.
The throughline is the method, not the subject. Immunai mapped immune cells. Jimini maps the space between therapy sessions. Both bets rest on the same conviction - that messy biological systems yield to rigorous models if you respect the domain enough to build for it specifically.
"The most important part of the entrepreneurial journey is who you're working with, because these things always change." Products pivot; co-founders are the constant. His closing advice to founders: "Work with people you love working with."
The defensible AI, he argues, is the one built deep into a specific domain - clinical supervision, evidence-based protocols, real workflows. A general model is a commodity. A model that knows the rules of therapy is a moat.
A scientist's habit: "Really guarding your time and making sure that you're spending time in the optimal way is really important." His teams take, in his telling, a more scientifically rigorous approach than most of the field.
He teaches founders at Stanford how to build AI companies - then goes home and builds one.
Voloch's pitch carries a number that makes the establishment flinch. He envisions a future where AI could "take 80-90% of the human effort out of providing care to some lower-acuity patients" - not to remove the clinician, but to stretch one therapist's reach across many more people who currently get nothing at all.
Read uncharitably, that sounds like automation come for the couch. Read the way Voloch means it, it is a distribution problem: there are not enough therapists, the ones who exist cost too much, and millions of people fall through. If software can carry the routine load under human supervision, the scarce human attention goes where it counts.
Whether the science holds is the open question, and Voloch - who won a best-thesis prize and then spent years modeling immune cells - seems to know it. He has built his company to answer it carefully rather than loudly.
It helps that he is not building alone. Jimini's founding team reads like a deliberate assembly of people who have already shipped hard things in health and software: Mark Jacobstein, a veteran of Guardant Health and Immunai, and Sahil Sud, who worked at Ribbon Health and Palantir. Three founders, overlapping résumés, a shared allergy to hype. That is the kind of lineup Voloch's own rulebook would predict, since he keeps insisting the team matters more than the idea.
The early backers agreed. Jimini's launch round drew Zetta Venture Partners, LionBird, PsyMed Ventures, BoxGroup, Arkitekt Ventures and SCB into a company that, at the time, was mostly a thesis about LLMs and a name for an assistant. By 2026 the total raised had climbed toward $25 million. For a founder who could have coasted on a unicorn, the second act is being financed by people betting he can do it twice.
A conversation on Business Trip / PsyMed Ventures about collaborative AI, clinical rigor, and why the once-a-week model is overdue for change.
"The AI secret that built a $1B+ healthtech company" - Voloch on Immunai, moats, and choosing co-founders.
The founding story of Jimini, the Genentech ambition, and his advice to healthcare entrepreneurs.