The idea behind Luventix didn't begin in a laboratory.

It began after a day of skiing.

Dejan Nenov was sitting around a dinner table with his friend and eventual Co-Founder Martin Martinov when something on television caught their attention. Dogs were sniffing urine samples. When a sample came from a cancer patient, the dogs sat down.

For two engineers, the obvious reaction wasn't simply amazement.

They wanted to know how it worked.

What was the dog's nose doing? It was collecting molecules evaporating from the sample. What was the dog's brain doing? It was recognizing a pattern and classifying what it detected.

Nenov and Martinov looked at each other and began asking a much bigger question.

Could technology reproduce that process?

Years later, that question sits at the heart of Luventix, where Nenov and his Co-Founders are developing a technology that combines metabolomics with artificial intelligence to find patterns associated with health conditions.

And if their bigger vision works, the humble urine sample could become far more informative than most of us imagine.

Nenov is Co-Founder of Luventix, alongside chief scientist Martin Martinov and Co-Founder George Holmes. Before healthcare, he spent his career mostly in software, moving through semiconductor design, cybersecurity, and energy.

In this episode of Lead with AI, host Dr. Tamara Nall speaks with Dejan Nenov about how Luventix is trying to replace biomarker guesswork with a full molecular picture, why a conversation with a pharmaceutical executive changed how he thinks about the business, and the personal moment that led him to write his own oath for engineers working in healthcare.

What 38,250 Data Points Could Say About Your Health  

During the conversation, Dr. Tamara Nall describes what Luventix creates as a kind of metabolic "digital twin," a framing Nenov runs with.

Forget the digital companion that phrase usually brings to mind. This is a snapshot of what is happening metabolically inside the body at a particular moment.

The premise begins with biology.

When the state of the body changes, metabolism can change with it. Cells behave differently and produce different metabolites. Some of those metabolites eventually appear in urine.f

Luventix measures those signals.

According to Nenov, processing a single urine sample through the instrument used by the company produces 38,250 data points.

AI enters the process by looking across that information for patterns associated with the presence or absence of a particular condition.

That distinction matters because traditional diagnostic development often begins by searching for individual biomarkers.

Researchers may spend years identifying the particular molecule, or combination of molecules, associated with a condition. Once those biomarkers are established, a diagnostic test still has to be developed and clinically evaluated.

Luventix is pursuing a different approach.

Rather than identifying one or several biomarkers first, its models look across the molecular information in the sample at once.

Nenov explains the difference with a dark room and a pair of dirty socks.

Imagine walking into that room with a flashlight. You search around and eventually find the dirty socks under the bed.

But there might be another sock sitting on top of the bed that your flashlight never reaches.

Luventix's approach, as Nenov describes it, is closer to turning on the lights.

The model gets to look across the available information instead of deciding beforehand which small part deserves attention.

One Sample Is Interesting. Dozens of Conditions Change the Equation.  

The most ambitious part of Luventix's vision goes past detecting one condition.

It's what could happen if the same sample could be used across many.

Nenov asks listeners to imagine giving one urine sample and eventually receiving probabilities for the presence or absence of dozens, or potentially even hundreds, of conditions.

He gives a more immediate example of roughly 50 conditions from a single laboratory run.

That possibility changes the economics of the idea.

The instrumentation behind the process is well understood, and Nenov says the technology Luventix uses to produce its metabolic digital twin is dramatically less expensive than something like genetic sequencing.

If one inexpensive sample can support many screening models, broad screening starts looking very different.

This is the part of the technology that excites Nenov most.

Healthcare costs are rising faster than GDP across much of the world. At the same time, earlier identification of disease often gives clinicians more options for intervention.

Nenov sees inexpensive, repeatable screening as one way technology could attack both sides of that problem.

His ambition reaches past a smarter diagnostic test.

It is making sophisticated screening economically realistic for far more people, regardless of socioeconomic status or geography.

Luventix is still developing that vision. Nenov makes clear in the conversation that the company does not yet have a finished product on the market and was wrapping up its first clinical trial.

But diagnostics are only one part of what caught the industry's attention.

Would a $50,000 Treatment Work? Here's How AI Could Tell You First  

One of Nenov's biggest "aha" moments came during a conversation with the CEO of a pharmaceutical company he had met about an hour earlier.

The executive immediately saw another possibility for the technology.

What if it could help indicate whether a patient was likely to respond to a particular treatment?

The question matters when the treatment costs tens of thousands of dollars.

Nenov points to some newer biologic therapies that can cost tens of thousands of dollars, with response rates he puts as low as 30 percent.

Think about what those numbers mean beyond a spreadsheet.

A patient undergoes an expensive treatment. A clinician invests time in a therapeutic strategy. An insurer pays the claim.

And the treatment may not work.

A tool that provides an indication of response or non-response beforehand could therefore matter to everyone involved.

The same concept has implications for clinical trials.

After researchers have assembled a trial cohort using their normal inclusion and exclusion criteria, Nenov says Luventix's technology could provide another layer of patient stratification based on the likelihood of response.

Better selection could make it easier to show whether a therapy actually reaches the endpoints of a study.

It also shows why the future of healthcare AI isn't only about diagnosing disease.

It may increasingly be about helping humans make better decisions before expensive, consequential actions are taken.

And that brings Nenov to the part of healthcare AI that technology alone cannot solve.

The Day "Faster" Stopped Meaning "Better"  

Nenov entered healthcare with the instincts of a software engineer.

Give an engineer a problem, and the natural response is to make the system work better.

So when his team encountered a medical testing process that could be improved, they did what engineers do.

They made it faster.

Much faster.

Nenov remembers presenting the system to clinical professionals. Samples could be integrated and processed efficiently, and results could reach the patient at essentially the same time as the doctor.

Cheaper. Faster. More efficient.

To an engineer, it sounded fantastic.

The clinicians didn't celebrate.

Instead, they asked him to imagine that the test revealed a patient was extremely sick and might have only months left to live.

Would you email them the result?

Would an automated system effectively tell someone to seek end-of-life care?

Nenov suddenly understood that an engineering success could become a human failure if nobody stopped to consider what happened on the other side of the screen.

That experience stayed with him.

He eventually wrote what he calls a software engineer's oath, modeled on the Hippocratic oath and beginning with the same fundamental responsibility to do no harm.

For Nenov, ethics in healthcare technology aren't an optional discussion to have once the product works.

They are part of the work.

AI Can Make the Recommendation. Who Owns the Decision?  

Near the end of his conversation with Dr. Tamara Nall on Lead with AI, Nenov is asked a deceptively difficult question.

If AI keeps improving at its current pace, which human skill becomes most valuable?

His answer is two words.

Judgment and responsibility.

AI agents, he argues, are becoming phenomenal at performing the tasks humans assign to them.

But deciding which tasks are worth doing remains a judgment call.

So does deciding whether the output is valuable.

Responsibility becomes even more important when those outputs influence someone's health.

Nenov is careful about the language here. He doesn't want to say that AI will diagnose the patient.

AI is the tool a clinician uses to help make the diagnosis.

The model could eventually be right 999 times out of 1,000 and Nenov would still argue that somebody has to remain responsible for the diagnosis, treatment plan, procedure, and care.

That may be one of the most important ideas in the entire conversation.

The more capable AI becomes, the easier it is to focus on what machines can now do.

Nenov is asking the opposite question.

What are humans still responsible for doing?

The Bigger AI Revolution May Be About Scale  

Nenov's predictions stretch well beyond healthcare.

He believes more than 90 percent of corporate AI projects fail to show real ROI. He points to physics-informed neural networks as an under-hyped area that could dramatically accelerate complex engineering calculations.

Then Dr. Nall asks him for his biggest prediction.

Nenov thinks about his own profession.

Software development, he says, is moving from an artisan model toward something resembling a factory. "Before current AI agentic technology, we were building software as artisans... Today, we are rapidly approaching the factory mode."

For decades, developers effectively built one custom car at a time. Agentic AI is beginning to change the economics of how that work gets produced.

His prediction?

Within five years, Nenov believes we will see the first company with roughly 10 employees generating a billion dollars in recognizable annual revenue.

Not a billion-dollar valuation.

A billion dollars in revenue.

It sounds extreme.

Then again, Luventix itself started with two engineers watching dogs sniff urine on television and asking if a machine could understand what the dogs already knew.

That is what makes Nenov's perspective on AI worth hearing.

The interesting question goes beyond what the technology can do today.

It's what happens when something that used to require years, millions of dollars, or hundreds of people suddenly requires a fraction of them.

In healthcare, that could mean extracting far more information from one ordinary sample.

In software, it could mean building billion-dollar businesses with teams small enough to fit around a conference table.

But in both cases, Nenov returns to the same boundary.

Capability does not remove responsibility.

If AI gives us better tools for seeing disease, selecting treatments, building software, and making decisions, humans still have to decide what should be built, how it should be used, and who answers for what happens next.

Quick Answers  

What is Luventix? Luventix is a diagnostics company using AI to analyze the full chemical makeup of a urine sample, rather than isolating a single biomarker, with the goal of screening for multiple health conditions from one sample.

How does Luventix's technology work? A gas chromatograph processes a urine sample and produces 38,250 data points. A machine learning model analyzes all of them together, looking for patterns associated with a given condition instead of searching for one known molecule.

Is Luventix's diagnostic test available to patients now? Not yet. Luventix is finishing its first clinical trial and developing lab-developed tests while seeking strategic partners to bring the technology to market at scale.

What is a companion diagnostic and why does it matter to Luventix? A companion diagnostic predicts whether a specific patient is likely to respond to a specific treatment before it's prescribed. Nenov sees this as one of the clearest paths to near-term value, particularly for expensive treatments with variable response rates.

Who is Dejan Nenov? Dejan Nenov is Co-Founder of Luventix. A software and electrical engineer by training, he spent years in Silicon Valley startups across semiconductor design, cybersecurity, and energy before moving into healthcare full time, where he co-founded Luventix with chief scientist Martin Martinov and George Holmes.

For pharmaceutical coSDmpanies, contract research organizations, and laboratories interested in a faster, less expensive path to screening or predicting treatment response, Luventix's work is underway at luventix.com.

For more conversations with the founders and leaders building the future of applied AI, subscribe to Lead with AI on your favorite podcast platform.

Follow or Subscribe to Lead with AI Podcast on your favorite platforms Website: leadwithaipodcast.com | Apple Podcasts: Lead-with-AI | Spotify: Lead with AI | YouTube: @LeadwithAIPodcast | Facebook: Lead with AI | Instagram: @LeadwithAIpodcast | TikTok: @LeadwithAIpodcast | Twitter (X): @LeadwithAI

Follow Dr. Tamara Nall LinkedIn: @TamaraNall | Website: TamaraNall.com | Email: Tamara@LeadwithAIPodcast.com

Follow Dejan Nenov (Co-Founder, Luventix) LinkedIn: @dejannenov | Website: luventix.com