"Originally, I wanted to go into mathematics" Mike said."There's a whole funny story here, but basically, I ​met Terence Tao and ​had this realization that I just wasn't quite smart enough to ​contribute meaningfully to the field"

What comes through as he speaks isn’t a lack of intelligence, but a desire for impact. Math was elegant, but too abstract ​—​​ medicine promised something tangible.

"Being a physician was really appealing because you could ​draw a straight line between your day’s work and the​​ difference ​you make ​in your patients’ lives” he explained. 

That instinct for impact showed up early. ​​In medical school, Mike started a student-run free clinic that, in its first year, cared for more than 1,000 uninsured patients.​​​​     ​​ 

“​Learning to ​pull​ ​talented people together to solve a problem felt really great,” ​​Mike said. “But this was around the time there were about 40 million uninsured people. ​​Although I couldn’t cut it in the field of mathematics, I was able to do some quick math: ​​​1,000 divided by 40 million​​ is actually ​not as much impact ​as you​’d hope for ​​relative to recruiting an entire medical school’s worth of volunteer​​s​​​​.” 

The numbers forced a realization: Clinical care alone has challenges scaling. That’s when Mike began to experiment with ​software development, eventually ​​launching ​​​FindCare​, a platform that connected uninsured patients to clinics with available capacity. It was an early signal of the path he’d subsequently follow — using data and technology to amplify healthcare’s reach far beyond what one physician could do alone. 

​​​Mike began his residency at New York-Presbyterian in 2014. By 2017, he was at Weill Cornell Medicine as an Assistant Professor of Medicine and Medical Director for Transformation, splitting his time between patient care and leading AI projects — including one on length of stay optimization that revealed something much bigger.

At the same time, hospitals were investing heavily in capturing exactly that complexity—working with clinical documentation improvement programs, improving medical coding, and pre-bill and post-bill review vendors.

"It was hard for me to reconcile that," Mike said. "On one hand, we were complaining about not capturing certain things. On the other hand, we had all these enormous investments into capturing these things." 

So he went digging. ​​"I sat in the hospital basement for a few months, wrote a bunch of code to try to look at the raw treatment data and see if there were things that, despite all of that upfront investment, were still leaking through or leaking around the edges."  

The timing was good. Just as he was uncovering gaps in clinical documentation, a new type of Artificial Intelligence​​ — ​​​transformer models — was emerging from research labs. Earlier tools w​​ere simple ​​statistical​​ or rule-based models​​​​. Transformers were different — they could read through huge amounts of information at once, connect the dots, and understand context. In practice, that meant they could digest clinical complexity in ways that felt more like human reasoning. 

Mike saw the convergence: widespread documentation gaps + better AI models = massive efficiency opportunity.

"That was the realization that if I'm finding all this​,​ and maybe these models will get more advanced, there's a way to get a bunch of really smart people to help me build better clinical AI,” he said. 

When Mike and his team first pitched SmarterDx, they focused on concurrent documentation improvement — reviewing charts while patients were still in the hospital. The reception was lukewarm.​     ​ 

This was where Mike leaned on a skill he credits to medicine: active listening. 

“When you’re asking about sensitive things like STDs or abuse or home safety, patients often aren’t going to flat-out give you the full answer. You have to be able to pick up on cues and tease it out a little bit.” 

He applied the same approach to customers. They weren’t saying “this doesn’t work” outright, but the real concerns showed through in subtle questions:

“How do I prove this to my CFO?” and “How do I know this is really working?” 

The insight was clear: ​Hospitals​ didn’t just want technology; they wanted proof. So SmarterDx pivoted. It became a final quality assurance layer, after all existing reviews, but before bills went out. 

“The big unlock was when we dropped the ability to do concurrent and just led with pre-bill,” Mike explained. “The clarity​ of our ROI​ finally broke through, and it started making sense to our customers.” 

That pivot set the stage for what SmarterDx would become. 

SmarterDx offers two complementary solutions, with the aim of enhancing the work of people, not outright replacing them: 

• SmarterPrebill combs through raw clinical data — labs, medications, vital signs — and flags likely documentation or coding gaps. Instead of replacing CDI nurses or coders, it gives them a targeted list of charts to validate. 

• SmarterDenials works on the back end, pulling the same clinical signals into evidence-based appeal letters. SmarterDx assembles the case, and humans confirm it. 

Hospitals, of course, need evidence that a solution is working. With tight margins, leaders are skeptical of anything that sounds like hype. Mike recalls one ​A​cademic ​Medical​ Center in the Northeast with five separate layers of documentation review. 

“They were rightfully skeptical” he said. “It’s like, ‘Hey, we’ve invested a ton. How is it that you’re able to find more?"

SmarterDx’s response was simple: work on contingency and only charge for validated findings. ​​That particular AMC saw 30 to 50 basis points of revenue improvement in a system with 120,000 discharges​.

Now, as part of SmarterTechnologies, ​Mike’s ultimate goal is to dramatically lower cost-to-collect so that health systems can put more resources back into what matters most: patient care and innovation.