What You Should Know

• UC Berkeley spinout Knit Health has officially launched out of stealth with $11.6M in seed funding co-led by Uncork Capital and Frist Cressey Ventures.
• The platform introduces a new AI framework called the Large Clinical Behavior Model (LCBM), which shifts away from traditional text-based large language models to focus on real-world clinical decision patterns.
• To train its behavioral world model, Knit Health leverages a massive data foundation consisting of Truveta EMR data from over 130 million patients across 30 major U.S. health systems.
• The underlying technology architecture synthesizes deep reinforcement learning, causal inference, and behavioral cloning to map and automate institutional coordination, including referral habits and scheduling practices.
• Operating as a deep infrastructure layer, the platform fine-tunes natively to an individual hospital’s specific capacity constraints and practice patterns to drive triage, patient flow, and discharge workflows.

---

Beyond the Textbox: Why Knit Health is Shifting Health IT from Language to “Clinical Behavior Models”

The rapid influx of artificial intelligence into American health systems has exposed a fundamental limitation in traditional machine learning architecture. The vast majority of healthcare AI platforms deployed today rely strictly on large language models (LLMs) trained on static text, clinical documentation, or published medical literature. While these text-centric tools excel at summarizing clinical notes or answering abstract medical questions, they are inherently blind to the operational realities of care delivery.

Traditional software routinely overlooks collective clinical intelligence—the highly complex, unwritten patterns of real-world decision-making, localized referral habits, scheduling practices, and informal institutional coordination that experienced clinicians use to navigate fractured hospital systems. This experiential knowledge is what ultimately drives patient outcomes, ensuring individuals are routed to the right care setting, at the exact right time, with the proper resources.

To capture this operational intuition and build a predictive infrastructure layer for health networks, University of California, Berkeley spinout Knit Health has officially launched out of stealth with an $11.6 million seed financing round. Co-led by Uncork Capital and Frist Cressey Ventures, with pre-seed backing from Moxxie Ventures and participation from Coalition Operators, the capital will be used to accelerate the development and systemwide deployment of Knit Health’s proprietary Large Clinical Behavior Model (LCBM).

Building a Behavioral World Model via Truveta EMR Data

Rather than attempting to train a better chatbot, Knit Health is constructing a behavioral world model designed to reflect how healthcare actually unfolds in physical practice. The company addresses the standard AI validation requirement by training its LCBM on a massive, high-fidelity data foundation: Truveta EMR data spanning more than 130 million patients across 30 major U.S. healthcare systems.

By moving past probabilistic text generation, the platform utilizes an advanced machine learning stack built on three core pillars:

• Deep Reinforcement Learning: Allowing the system to evaluate and optimize complex multi-step operational routing and care tracking loops over time.
• Causal Inference: Enabling the algorithm to distinguish between mere statistical correlations and true cause-and-effect patterns within longitudinal patient journeys.
• Behavioral Cloning: Directly mimicking the high-value decision sequences executed by seasoned clinicians when they manage capacity constraints, coordinate discharges, or execute specialty referrals.

Jonathan Kolstad, Co-founder and CEO of Knit Health, noted that much of what matters most in medicine cannot be found in a textbook; it is learned through decades of localized experience. Knit Health captures these real-world decisions, transforming collective experience into an objective asset that continuously improves systemwide efficiency.

Localized Optimization: Shifting the Axis of Medtech Competition

A common failure mode for generic generative AI models is their inability to adapt to the unique, hyper-local operational environments of individual facilities. Knit Health addresses this integration bottleneck by operating as an enterprise infrastructure layer that customizes itself to the specific practice patterns, regional referral dynamics, and physical bed capacity constraints of each partnering health system.

Once fine-tuned, the LCBM sits silently beneath the hospital’s core operational systems, injecting data-driven intelligence into critical, high-friction administrative tasks. This includes automating triage classification, predicting optimal discharge windows, managing care team allocations, and smoothing patient flow throughout the facility.

Navid Farzad, Managing Partner at Frist Cressey Ventures, highlighted that the primary challenge in modern healthcare is not identifying what excellent care looks like, but delivering it consistently for every single patient. Embedding this intelligence directly into the clinical workflow allows frontline providers to make highly calibrated decisions faster and more predictably.