Nishith
Khandwala

There's a specific kind of frustration that only happens inside academic labs. You build something beautiful - a model that outperforms radiologists at detecting coronary calcium, validated across thousands of scans, published at a conference. And then it just... sits there. Nishith Khandwala sat with that frustration long enough to do something about it.

He came to Stanford as a computer science student in 2013. His first instinct wasn't medicine - it was chess. His 2015 project ConvChess applied convolutional neural networks to predict chess moves. Pure computer science, no stethoscopes in sight. Then Stanford physicians walked into his lab. They had patients. They had data. They had problems that algorithms could solve - but no way to get those algorithms out of notebooks and into clinical systems.

The pivot was personal before it was professional. His father had a heart attack. Nishith had literally built an AI that could detect coronary artery calcium on a routine CT scan - a predictor of cardiac risk that most patients never get screened for. And yet that algorithm wasn't deployed at the hospital where his father was treated. It wasn't deployed at most hospitals. It was a paper. He decided the problem wasn't the algorithm. The problem was the infrastructure - the pipes and plumbing that would need to exist before any algorithm could reach any patient.

So in 2019, Nishith and co-founder David Eng joined Y Combinator's Winter 2019 batch and started Bunkerhill Health. The name invokes the spirit of a contested battle - fitting for a company fighting to rewire how healthcare systems absorb new technology.

The original Bunkerhill model was a consortium - partnering with researchers at over 20 academic medical centers to help them develop, validate, and deploy AI algorithms. It answered the question: who will make FDA-cleared algorithms available? But it was answering only half the problem. Even FDA-cleared algorithms pile up unused inside health systems that lack the workflow infrastructure to act on their outputs.

Carebricks launched publicly in late 2023 and has expanded rapidly. Health systems using the platform include UTMB Health, HCA Florida Healthcare, MedStar Health, Endeavor Health, WVU Medicine, Stamford Health, and McLaren Health Care. UTMB now runs Bunkerhill workflows across 7+ clinical domains - the kind of enterprise-scale deployment that most healthcare AI startups only put in pitch decks.

The platform spans three main surfaces: addressing actionable findings (detecting and escalating missed care opportunities from radiology reports, lab results, and clinical data), streamlining prior authorization (assembling and submitting authorization packets), and improving case-mix accuracy (enhancing documentation and coding for complex conditions). Underneath all of it is Carebricks Chat, launched in early 2025 - conversational AI that gives clinicians access to the entire patient record.

FDA clearance is the proof-of-work for medical AI. Getting one is hard. Getting seven is a pattern. Bunkerhill's cleared algorithms span cardiology, radiology, and pulmonology:

There's a number nobody advertises: the average time from a published clinical AI model to live patient use is measured in years, not months. Papers pile up. Hospitals lack the infrastructure to integrate new tools into their EHR workflows. Clinicians lack the time to change behavior. Compliance and IT requirements create choke points. The algorithm that could change outcomes for a patient sitting in a waiting room right now may be waiting for a procurement cycle to complete.

Nishith identified this bottleneck while still a researcher. The frustration crystallized in a phrase he said in an interview: "This is why academia sucks. This is why people burn out." Not an indictment of the researchers - an indictment of the missing layer between insight and action. He left to build that layer.

Bunkerhill's vision - compressing the time from clinical AI idea to live deployment from years to hours - isn't rhetorical. The Carebricks platform is built as modular infrastructure: health systems can configure and deploy new AI workflows without starting from scratch each time. It's a platform strategy applied to a space that has historically operated project-by-project.

The company is SOC 2 Type II certified, ISO 27001 certified, and HIPAA compliant - the table stakes for enterprise healthcare. These are not marketing bullets; they're the price of admission for any conversation with an enterprise health system's IT and compliance teams.

UTMB Health runs Bunkerhill workflows across 7+ clinical domains - a level of enterprise integration that validates the platform model over point-solution deployments.