The pitch sounds almost too ambitious: artificial intelligence that doesn't just read medical charts, but learns to anticipate the thousands of micro-decisions that keep a hospital running—when to discharge a patient, which specialist to page, how to prevent an emergency room from buckling under demand.
Yet that's exactly what Knit Health is attempting. The San Francisco startup, founded in 2025, announced Tuesday that it had closed an $11.6 million seed round to build what it's calling a Large Clinical Behavior Model. Unlike the documentation-focused AI tools that have flooded healthcare in recent months, Knit's system is trained on something messier and more valuable: the actual patterns of how physicians route, schedule, and coordinate care in the wild.
Uncork Capital and Frist Cressey Ventures co-led the May 12 funding round. Moxxie Ventures, which had backed the company earlier in a pre-seed alongside Coalition Operators, also participated.
The bet reflects a broader shift in healthcare AI investment—away from glorified chatbots and toward systems that tackle operational chaos. For health system executives, the appeal is immediate. Patient flow bottlenecks cost hospitals billions annually in lost revenue and strained capacity. If an algorithm could predict admissions, flag discharge-ready patients, or intelligently allocate scarce specialists, the efficiency gains would be substantial.
But building such a model requires something most AI companies don't have: access to granular, high-fidelity data about how care actually unfolds.
The Data Engine
On the same day it announced funding, Knit unveiled a partnership with Truveta, the health system-owned data consortium that has quietly become one of the most valuable repositories of real-world clinical information in the country. Truveta's dataset—refreshed daily and covering more than 130 million patients across over 30 U.S. health systems—includes not just billing codes and lab results, but clinical notes, imaging metadata, mortality outcomes, and social determinants of health.
For Knit, the collaboration offers something rarer than scale: provenance. The data comes with documented lineage and audit trails, critical for a model designed to influence operational decisions in environments where mistakes carry real consequences.
"This goes beyond textbook medicine," the company said in its announcement, a pointed jab at the wave of large language models trained primarily on published literature.
What Knit is attempting instead is more complex. Its system uses deep reinforcement learning, causal inference, and a technique called behavioral cloning to reverse-engineer the implicit logic behind clinical workflows. How do experienced attending physicians decide when a patient is stable enough for discharge? What signals trigger a call to a cardiologist rather than a hospitalist? The model isn't trying to replace clinical judgment—it's trying to surface the patterns that often go undocumented.
Whether that approach can scale across the bewildering variety of hospital systems, each with its own quirks and workflows, remains an open question.
Academic Pedigree
The founding team brings considerable research credibility. Jonathan Kolstad, co-founder and CEO, holds the Henry J. Kaiser Chair at UC Berkeley's Haas School of Business and is a research associate at the National Bureau of Economic Research, specializing in health economics. He previously co-founded Picwell, where he served as chief data scientist.
Co-founder and Chief Scientist Maya L. Petersen is a professor of biostatistics and epidemiology at Berkeley and co-directs both the UCSF-UC Berkeley Joint Program in Computational Precision Health and Berkeley's Center for Targeted Machine Learning and Causal Inference. Her work has centered on using causal inference to extract actionable insights from complex observational data—precisely the challenge Knit faces.
The broader leadership roster includes co-founders Jonas Knecht (Co-CTO) and Ted Robertson (COO), along with Co-CTO Anshul Amar, Chief Product Officer Sophie Pinkard, Head of Data Science Anna Zink, and Chief Growth Officer Midori Uehara.
The Operational Frontier
Knit plans to deploy its seed capital toward refining the model and embedding it within health systems. Initial use cases focus on predictive patient flow: anticipating emergency department admissions, identifying discharge-ready patients, flagging candidates for hospital-at-home programs, and forecasting transfer needs. The company is also targeting intelligent specialist routing and care team allocation—problems that often devolve into frantic paging and guesswork during peak census periods.
The company hasn't disclosed specific health system customers, though its focus areas align neatly with operational pain points that have driven hospitals to move AI pilots into live production. Knit positions its technology as an "intelligence layer" designed to integrate across existing systems, with HIPAA-aligned safeguards, encryption, and role-based access controls baked in.
For Frist Cressey Ventures, which closed a $425 million Fund IV in February and now manages approximately $1 billion in assets, the investment fits a deliberate thesis around operational AI in care delivery. Uncork Capital, the seed-stage firm, characterized the bet as backing "clinical intelligence of the future" in a May 12 blog post—though the future they're describing is one where hospitals run less on instinct and more on algorithmic foresight.
Whether clinicians will trust that foresight enough to act on it is another matter entirely. The history of healthcare AI is littered with promising models that worked beautifully in retrospective datasets but stumbled when introduced to the messy reality of bedside decision-making. Knit's challenge will be proving its system can handle not just the patterns it's been trained on, but the exceptions that define hospital medicine.
