We started on the other end of this problem
MTN stands for “Medical Timeseries Networks”. We built the MTN Guide to solve our own data engineering problems.
The wall we hit
We came to this from computational neuroscience and clinical medicine, working on models that find patterns in physiological and clinical data as they unfold over time. We built the pipelines, the wearable integrations, the dashboards.
And we kept losing the same months to the same work. Not to the models, but to figuring out what the data meant. Which of five timestamp columns marked when the encounter actually happened. Whether this system’s visit was the same thing as that system’s encounter. Why a report broke after a vendor update that changed three field names.
Our ML scientists were doing data archaeology. Every new project started the work over. And every answer lived in one engineer’s head until they left.
So we built the layer underneath
MTN Guide is an agentic data engineer. It reads schemas, documentation, and API definitions, and builds MTN FieldMap: an evidence-backed map of what an organization’s data means, then maintains that map as systems change. It is the thing we needed before we could do any of the work we actually set out to do.
Two convictions shaped it. The first is that a map has to be honest about its own uncertainty: a confident wrong answer is worse than no answer, so Guide records the evidence under every claim and refuses plainly when the evidence isn’t there. The second is that you cannot merge your way out of this. Two departments genuinely disagree about what a customer is, and a single unified model just hides the disagreement. So we map instead, recording where systems agree, where they overlap, and under what conditions.
It turned out this was not our problem. It was everyone’s.
Built for the messy case
Partial, inconsistent, and drifting data isn’t an edge case we handle. It’s the condition the product assumes, and the reason it exists.
Built for High-Trust
The map is built from how your systems are shaped, so the sensitive part never has to move. Security review gets a lot shorter.
Why now
For decades, understanding a data estate required expensive humans reading it one system at a time. That was affordable for large enterprises and out of reach for everyone else. In late 2025 coding agents became trustworthy enough to do the reading, and the economics inverted.
The urgency arrived at the same moment. Messy data used to be a hindrance that slowed a company down. Handed to an unchecked agent, it now lies with confidence.
Understanding a data estate meant hiring people to read it, one system at a time. Only large enterprises could afford the climb.
Machines got good enough to read structure and hold their reasoning to evidence.
The read now takes hours instead of quarters, which puts a trusted foundation within reach of operators who were priced out of one.
Technical leadership
An unusual combination: ML scientists who understand deployment constraints, physicians who understand data infrastructure, and engineers who turn it into product. Our work has been published in Nature journals, PNAS, JMIR, Chest, PLoS Computational Biology, The Royal Society, and other leading venues.
Co-Founder & CEO
A machine learning scientist and software builder who has wrestled the data engineering problem for 17 years. Trained in medicine at Colorado and computational neuroscience at Harvard, Stanford, NYU, and Yale. He co-founded MTN in 2023 to address the under-utilization of data in healthcare. Separately, he leads the Medical Machine Intelligence (M²Int) Lab at the University of Utah, an academic research group developing AI for clinical applications. Prior service in U.S. and Colorado health policy, and on the University of Utah IRB, shapes MTN's governance posture.
Enterprise AI Advisor
Head of Medical Data & AI at Sanoptis, one of the largest ophthalmology networks in Europe. PhD and postdoctoral research at Columbia University in computational ML. Previously a Deep Learning Research Engineer at DeepLife, training foundational models on genomic and biometric data. Investigator with the M²Int Lab. Aligns MTN's products with the integration and scalability needs of M&A-driven enterprises.
“We were tired of spending more time on data plumbing than on actual science. So we built a system that could handle the integration complexity for us. Turns out, that system is exactly what a lot of other organizations need.”
— Warren Pettine, Co-Founder & CEO
Vetted validation
Most of what MTN has built was paid for by research funding we had to win in open competition. Getting that funding meant passing peer review: independent scientists, physicians, and engineers with no connection to the company take the technical approach apart and judge whether it will actually work. It is designed to be hard to pass, and most applications don’t.
MTN has been through it three times, in front of three different panels.
$2.34M. Funded on first submission.
Won to build time-series models.
Finalist. Tested on Army data.
We are grateful to the Nucleus Institute, whose UTIF grant and SBIR application support helped make the NIH award possible. Federal funding supports this research; it is not an endorsement of MTN or its products by NIH, NIA, or the U.S. Army.
Our mission
Healthcare generates an extraordinary record of what happens to people over time, and most of it goes unused because nobody can say with confidence what it means. MTN exists to close that gap: to make fragmented, irregular, sequential data usable, without asking organizations to hand over the data itself or pretend to a certainty they don’t have.
We’re building the layer that makes the rest possible.
Want to learn more?
Whether you're deploying clinical AI, integrating after an acquisition, or just trying to find out what your own data means, we'd like to hear from you.