For 15 years, Dr. Mario Martone’s daily work revolved around the universe’s most profound mysteries. Armed with a double PhD, he operated at the cutting edge of theoretical physics, studying black holes and the mathematical complexities of string theory.

Today, you won’t find him pondering the cosmos. Instead, Dr. Martone serves as the Head of Applied Research at Elise AI, a conversational AI company focused on transforming essential industries like housing and healthcare.

In a recent interview with MTS, Martone detailed his extraordinary career pivot, revealing why he traded the slow, abstract world of academia for the lightning-fast pace of applied machine learning—and why he believes the challenges he faces now are even harder than string theory.

The Need for Speed and Real-World Impact

In theoretical physics, progress is measured in years, if not decades. AI, by contrast, is a field that reinvents itself almost daily. For Martone, this relentless pace was a major draw, but it was the desire for tangible impact that ultimately drove his career change.

While he could have joined a frontier AI research lab, Martone recognized that the work there is often just as abstract as theoretical physics. Elise AI presented a unique opportunity: applying bleeding-edge intelligence to industries that desperately need it.

“The decisions that the LLMs make for Elise, for our products, they’re not just some random things that appear on the screen,” Martone explains. “They drive impacts on people’s lives, and that could be very positive or very negative. That’s the thing that really excites me.”

The Stakes: Why 90% Accuracy Simply Isn’t Good Enough

One of the biggest culture shocks in transitioning from academia to enterprise AI is the definition of “success.”

In academic papers, a language model achieving a 90% success rate on a benchmark is often celebrated as a breakthrough. But Elise AI handles roughly 5 million phone calls a month—or 60 million a year. At that scale, a 10% failure rate translates to 6 million botched interactions in essential sectors like healthcare and housing.

“You can really see the difference between people who are thinking about this problem theoretically and people who actually pay the consequences of those mistakes,” Martone notes. To bridge this gap, Martone has applied the rigorous scientific method to product design, building robust evaluation loops (“evals”) that allow a small team to systematically monitor the quality and impact of millions of automated conversations.

Beyond the LLM Hype: Mixing AI Modalities

A common pitfall in today’s tech landscape is the assumption that Large Language Models (LLMs) are the silver bullet for every problem. Martone’s role allows him a bird’s-eye view of the entire AI ecosystem, and his approach is highly pragmatic.

“People who have joined AI very recently, they just have one solution for everything: LLMs,” he says. But when you look beneath the surface, LLMs struggle heavily with deterministic tasks and probability outputs.

To solve this, Elise AI employs a full spectrum of tools. For conversational voice latency—which requires hundreds of operations in less than a second and a half—they rely on highly optimized proprietary stacks. For predicting lease renewals based on vast historical data, they skip the LLMs entirely in favor of elaborate, custom-built traditional neural networks.

By running immediate, internal benchmarks the moment a new open-source model drops, Martone’s team can rapidly fine-tune and push the best possible tools into production without falling for social media hype.

What’s Next: The Mission to Automate Affordable Housing

Elise AI has already fully automated large chunks of the leasing process. A prospective renter can discover an apartment, book a tour, receive an entry code, and be guided through the physical space by an AI—all without human intervention. They are also automating maintenance routing, leaving only the physical labor of the repair to human hands.

But Martone’s team is now setting its sights on an even larger problem: affordable housing.

Housing and healthcare combined account for roughly 40% of the US GDP. Within that, affordable housing is a massively complex, under-digitized sector serving the most vulnerable populations. By aggressively introducing automation into this space, Martone believes they can drastically reduce the operating costs of affordable housing, directly and immediately improving the lives of those who need it most.

A Complexity Greater Than the Cosmos

To solve these problems, Elise AI is hunting for “unicorns”—top-tier developers and academic researchers who possess a profound depth of thought, paired with a burning desire to execute in the real world.

For anyone wondering if moving from theoretical physics to property management AI is a step down in intellectual rigor, Martone has a definitive answer.

“I did not expect that, but even after coming from string theory, I would say the complexity of the problems we’re working on now is probably higher than it was before.”

Based on Dr. Mario Martone’s interview on the MTS Podcast. You can watch the full episode here.