What if the next major leap in understanding the universe didn’t come from a billion-dollar space agency or a tenured physicist—but from a teenager in a dark room debugging code?

In the vast silence of space, countless stars flicker, explode, or vanish without notice. For years, some of these changes went unrecorded, hidden deep within over 200 terabytes of dusty infrared data collected by NASA’s NEOWISE telescope. It took an 18-year-old high school student, Matteo Paz, to teach a machine how to see what no one else had seen—1.5 million new cosmic objects, each one a potential clue to the nature of time, gravity, and the fabric of reality itself.

With a self-built AI model, Paz didn’t just sift data—he transformed it. And in doing so, he challenged assumptions not only about space, but about who gets to make meaningful discoveries. His story isn’t just about algorithms or telescopes—it’s about perception, possibility, and the overlooked intelligence in plain sight.

Here’s how one student cracked open the sky.

From Intern Task to Astronomical Breakthrough

When Matteo Paz arrived at Caltech as a high school intern, he was handed what most would consider scientific grunt work: a mountain of raw, unlabeled data from NASA’s retired NEOWISE infrared space telescope. The dataset—200 terabytes in total—contained over 200 billion rows of observations spanning a decade of sky surveys. It was unfiltered, unclassified, and, for the most part, untouched.

This kind of task isn’t uncommon in research institutions. Interns are often brought in to handle data organization or background analysis—important but rarely transformative work. But instead of approaching the assignment as a temporary duty, Paz saw it as a gateway. He asked a simple but powerful question: What if I could teach a machine to find what humans don’t have time to see?

With a foundation in formal mathematics and computer science from his district’s advanced math program—where students complete college-level calculus before high school—Paz built a custom machine-learning model capable of analyzing the NEOWISE data for signs of variability in light. Variable objects—such as black holes, eclipsing binary stars, and supernovae—leave behind faint patterns in light intensity that change over time. These subtle shifts, often buried in noise, were exactly what Paz trained his AI to detect.

The Student-Built AI That Revolutionized Infrared Astronomy

His model, later named VARnet, incorporated advanced mathematical techniques to clean, compress, and analyze the data without erasing meaningful fluctuations. Among the most novel components was a method called the Finite-Embedding Fourier Transform, which helped extract periodic patterns from noisy measurements—critical in spotting objects that pulse or flicker unpredictably.

The initial plan had been modest: analyze a narrow slice of the sky for a handful of variable stars. But the algorithm showed early promise, and Paz decided to scale it—dividing the entire sky dataset into 13,000 segments and processing each one. What followed was an exponential leap: from a limited exercise in pattern recognition to a comprehensive, AI-driven sky survey.

By the end of the project, Paz had reviewed data on over 450 million celestial objects. Out of these, 1.9 million were flagged as variable, and 1.5 million had never been cataloged before.

In astronomy, such a yield is extraordinary—not just because of the volume, but because these are dynamic, energetic objects that could rewrite parts of our understanding of how the universe evolves. Even more remarkable is that this wasn’t a multi-year PhD project. It began as a summer internship.

Decoding the Cosmos: How Data Shapes Our View of the Universe

Astronomical data is often described as overwhelming—not just in scale, but in complexity. The NEOWISE mission, originally launched to detect near-Earth asteroids, quietly amassed a staggering 200 terabytes of infrared observations over more than a decade. That’s the equivalent of roughly 40,000 high-definition movies’ worth of information, not about our planet, but about the sky above us. Most of it sat largely unexamined, simply because it was too big, too dense, and too chaotic to process by hand.

Enter machine learning—not as a novelty, but as a necessity. Matteo Paz’s innovation wasn’t in inventing a new telescope or mathematical theorem, but in architecting a system that could listen to the rhythms hidden in this data. His algorithm, VARnet, was tailored not to look for known objects, but to detect change itself—the subtle flickers, fades, and pulses that indicate a star is dying, a black hole is feasting, or a pair of stars is eclipsing each other in gravitational lockstep.

Paz built his system in layers. The first step was “denoising” the raw measurements, preserving meaningful short-term changes while filtering out background interference. The second was applying the Finite-Embedding Fourier Transform, a method designed to extract recurring patterns—essentially giving the algorithm a way to “feel out” time-based variation. The result was an architecture that didn’t just flag anomalies, but began to classify them: black holes, supernovae, dual-star systems, and more.

The Grueling Path to a New Universe

But innovation doesn’t always unfold cleanly. Paz spent long hours troubleshooting bugs, often alone in a dark room, scribbling equations on whiteboards and running tests that failed more often than they succeeded. What drove him wasn’t the prestige of the science fair or the promise of a prize—it was the realization that the algorithm was beginning to reveal structure in what had seemed like noise.

He divided the full dataset into 13,000 manageable chunks, feeding each through VARnet. And in this repetitive, almost meditative process, the sky itself began to change. In constellation after constellation, the algorithm uncovered variable objects that had never been recorded. It didn’t just detect light; it mapped transformation.

The culmination of his efforts is a dataset called VarWISE, a catalog of nearly two million variable objects now being used by astronomers to study everything from distant binary systems to exoplanet mass. It is, effectively, a new map of the night sky—not static and serene, but dynamic, volatile, and alive.

In a field where big discoveries often come from new telescopes or billion-dollar missions, Paz’s work is a reminder of something more elemental: the future of astronomy may not depend solely on collecting more data, but on how we choose to look at the data we already have. The cosmos has been speaking all along. We just needed someone to listen differently.

Hubble’s Dilemma and the Catalog That Could Solve It

Matteo Paz’s discovery of 1.5 million celestial objects offers more than a statistical achievement; it provides vital insights into the fundamental workings of the universe. His catalog focuses on variable objects—celestial bodies whose brightness fluctuates over time, revealing critical astrophysical processes. A black hole may flare as it consumes nearby matter, or a supernova can outshine entire galaxies before fading. These variations are not mere curiosities; they are essential tools for understanding cosmic events.

Paz’s work has enabled scientists to uncover previously hidden phenomena. Researchers at Caltech are using his VarWISE catalog to identify eclipsing binaries, which are key to calculating stellar masses and distances. This foundational research is crucial for refining our models of stellar evolution and understanding the conditions necessary for planet formation.

Paz’s discoveries also touch on one of cosmology’s greatest debates—the rate at which the universe is expanding, known as the Hubble tension. By identifying variable objects like Cepheid variables and supernovae, which serve as cosmic yardsticks, Paz’s catalog could help resolve this discrepancy. His data may reveal new layers of complexity in current models of the universe’s expansion.

Furthermore, some of the objects Paz identified don’t fit known categories, pointing to potentially new classes of celestial behavior or even new physics. These “quirky, mysterious objects” challenge our current understanding and push the boundaries of what we know. In a field that thrives on both precision and wonder, Paz’s catalog not only answers questions but raises new ones, opening the door to future discoveries that could reshape our view of the cosmos.

Mentorship, Grit, and the Human Side of Genius

Matteo Paz didn’t walk into Caltech with a complete algorithm or a grand vision. What he did bring was a deep-rooted curiosity, sharpened by years of self-driven learning and a love for astronomy that began in grade school, when his mother took him to public stargazing lectures. That early spark, combined with a solid foundation in math and computer science through Pasadena’s advanced Math Academy, laid the groundwork. But it was the environment around him—the access, the mentors, the space to try and fail—that allowed that spark to catch fire.

Central to this story is Davy Kirkpatrick, a senior scientist at Caltech’s Infrared Processing and Analysis Center (IPAC), who had spent years mentoring students, often those overlooked by traditional academic pipelines. When Paz expressed a desire to publish a scientific paper—a goal far beyond the scope of his six-week summer program—Kirkpatrick didn’t laugh it off. He leaned in. He offered guidance, connections, and trust.

Kirkpatrick’s own path—growing up in rural Tennessee and being guided by a high school science teacher who saw his potential—informed his commitment to passing that support forward. “If I see their potential,” he said, “I want to make sure that they are reaching it. I’ll do whatever I can to help them out.”

That kind of mentorship isn’t just about academic advice; it’s about belief. It creates the psychological space where ambitious thinking becomes possible. And Paz responded with grit. In interviews, he recalls long nights alone in a lab room, eyes burning from screen light, deep in code and equations. He speaks of the Eaton Fire that forced his family to evacuate during the project, of bugs in his system that took days to fix, of the mental toll of uncertainty.

And yet, he persisted—not for a grade or competition result, but because the process itself was enthralling. He described visualizing the plotted sky from his discoveries as “very beautiful,” not in poetic abstraction, but in the tangible, mathematical elegance of what the data revealed.

Even as Paz moved from intern to part-time Caltech employee, he began mentoring other high school students, echoing the support he had received. That shift—from learner to collaborator—is a powerful indicator of how science grows not only through findings, but through relationships. Knowledge is never purely individual. It is built across generations, through trust, patience, and shared vision.

How a Teen Expanded Our Cosmic Consciousness

A teenager discovering 1.5 million unseen cosmic objects—by teaching a machine to notice what others overlooked—challenges more than scientific assumptions. It asks: What else are we missing because we haven’t trained ourselves, or our tools, to see it? On the surface, Matteo Paz’s story is about AI and astrophysics. But beneath that lies a deeper theme—the relationship between human curiosity and the unknown. His algorithm didn’t just analyze data; it tuned into a cosmic intelligence, revealing itself through patterns too subtle for human eyes.

These objects—supernovae, black holes, eclipsing stars—aren’t static. They’re dynamic processes unfolding across time and space, now witnessed through machine learning and human intention. This merging of computation and cosmic inquiry reflects a profound truth: intelligence, whether in code, stars, or consciousness, transcends boundaries. The same mind that ponders everyday problems can also ask, How fast is the universe expanding? Why do stars behave strangely?

There’s a spiritual humility in Paz’s achievement. The universe rewarded his curiosity with insight, suggesting a deeper reciprocity—one where science, practiced with reverence, becomes an act of communion. In revealing hidden cosmic objects, Paz also revealed something in us: the enduring desire not to conquer mystery, but to engage with it meaningfully.

Featured Image from Website @SocietyForScience

Loading...