A High Schooler's Open-Source AI Just Uncovered 1.5 Million Hidden Cosmic Objects

For decades, the scale of the universe has been matched only by the sheer scale of the data required to observe it. Institutional supercomputers and entire university departments have traditionally held a monopoly on processing the petabytes of information beamed back by space telescopes. But in early 2026, that paradigm shifted dramatically when Matteo Paz, a student at Pasadena High School, uncovered a cosmic treasure trove containing 1.5 million previously unknown space objects.

Paz's journey began in the summer of 2022 when he joined the Planet Finder Academy, a program designed to immerse students in real-world astronomical challenges. Working under the mentorship of Caltech scientist Davy Kirkpatrick at the Infrared Processing and Analysis Center (IPAC), Paz was granted access to a massive, publicly available archive from NASA's NEOWISE telescope.[2]

Launched originally in 2009, the NEOWISE mission spent over a decade collecting full-sky infrared observations, primarily hunting for near-Earth asteroids. By the time Paz began his analysis, the dataset had swelled to an almost incomprehensible size, capturing countless celestial objects and distant phenomena across the dark sky.

The archive contained nearly 200 billion rows of measurements. The challenge was not a lack of data, but an overwhelming abundance of it. The variations in the infrared signals were often too subtle, too slow, or too brief for human eyes or standard algorithmic scanning to detect, leaving millions of dynamic events buried in the noise.

Rather than relying on standard image recognition models, Paz built a custom machine learning framework focused entirely on time-domain signal processing. He recognized that the key to finding hidden objects wasn't just looking at static images, but analyzing how the light from specific coordinates changed over a decade.[1]

To achieve this, the teenager's model utilized Fourier transforms and wavelet analysis—mathematical tools highly effective at studying time-based signals. These techniques allowed the AI to break down complex infrared light curves into distinct frequencies, revealing faint, localized variations that NEOWISE's standard sampling had missed.[1]

The results were staggering. The algorithm flagged 1.5 million previously invisible phenomena, ranging from distant brown dwarfs and variable stars to slow-moving anomalies that defied immediate classification. Objects that changed so slowly they appeared static to older algorithms were suddenly pulled into sharp focus.[1][2]

While the initial discovery was made on a consumer laptop, validating the findings required rigorous institutional backing. Paz and his mentors at Caltech spent months cross-referencing the AI's output against known astronomical catalogs, overcoming initial skepticism from the broader scientific community about the reliability of a student-built model.[2]

The validation culminated in a peer-reviewed publication in The Astronomical Journal, earning formal praise from NASA leadership. The paper detailed how the wavelet-based machine learning approach successfully depolymerized the massive data archive into actionable, highly accurate astronomical targets.[1]

The impact of the catalog is already being felt at the highest levels of professional astronomy. In February 2026, the Vera C. Rubin Observatory in Chile—a flagship facility designed to conduct a decade-long survey of the dynamic universe—integrated Paz's data into its real-time alert system, using the teenager's discoveries to guide its multi-billion-dollar lenses.[3]

Beyond the specific astronomical discoveries, the project highlights a massive shift in modern science: the democratization of astrophysics. By leveraging open-source AI libraries and publicly accessible government datasets, independent researchers can now perform analyses that once required elite institutional funding.[4]

This transition from mainframe supercomputers to consumer hardware is rewriting the rules of scientific engagement. It proves that the barrier to entry for paradigm-shifting research is no longer hardware or funding, but rather curiosity, mathematical fluency, and a fresh perspective on existing data.

The underlying architecture of Paz's model also holds promise far beyond the stars. Because the algorithm is fundamentally designed to analyze subtle, time-based variations in massive datasets, researchers are already exploring how it could be adapted for environmental monitoring, high-frequency finance, and neuroscience.[4]

Ultimately, the mapping of the dark sky by a high school student serves as a powerful reminder for the global scientific community. As AI tools become increasingly accessible, the next major leap in human knowledge might not emerge from a corporate lab, but from a teenager with an internet connection and the audacity to look closer.[4]