High School Senior Uses AI to Uncover 1.5 Million Hidden Celestial Objects in NASA Archives

Matteo Paz, an 18-year-old from Pasadena, California, has achieved what entire university departments strive for. Using a custom-built artificial intelligence algorithm, the high school senior sifted through a decade of archived NASA data to uncover 1.5 million previously unidentified celestial objects. The unprecedented discovery has sent ripples through the astrophysics community, proving that the next era of space exploration is being driven by a combination of youthful curiosity and advanced machine learning.

The sheer scale of the discovery is staggering. The newly identified phenomena include flickering quasars, eclipsing binary stars, and distant supernovae that had remained hidden in plain sight. By automating the detection of subtle infrared light variations, Paz transformed a dormant, overwhelming database into a highly searchable treasure trove of cosmic activity, expanding the known catalog of variable space objects by orders of magnitude.[2]

The journey began in the summer of 2022 when Paz enrolled in the Planet Finder Academy at the California Institute of Technology (Caltech). The program, designed to immerse high school students in real-world astronomical challenges, paired him with Dr. J. Davy Kirkpatrick, a senior scientist at Caltech's Infrared Processing and Analysis Center (IPAC). Under Kirkpatrick's mentorship, Paz was given the opportunity to tackle genuine scientific bottlenecks.[1][3]

Kirkpatrick introduced Paz to the archives of the Near-Earth Object Wide-field Infrared Survey Explorer (NEOWISE). Originally launched in 2009 to track asteroids and comets, the space telescope spent more than a decade scanning the entire sky in infrared light. Because it repeatedly observed the same patches of space, NEOWISE inadvertently recorded how the brightness of distant objects changed over time, capturing the shifting heat signatures of rare celestial events.[1]

However, the dataset was overwhelmingly massive. It contained nearly 200 billion individual rows of measurements, representing almost 200 terabytes of raw information. Traditional manual analysis or standard software tools were entirely inadequate for a dataset of this magnitude, leaving the vast majority of the archive untouched and its secrets locked away in a sea of numbers.[4]

Rather than analyzing a small sample, Paz decided to build a machine-learning pipeline capable of processing the entire catalog. Drawing on his advanced mathematics background from the Pasadena Unified School District, he developed an AI model dubbed VARnet. The goal was to teach a computer to scan the full archive for changes in brightness over time, turning raw data into actionable insights.[3]

The algorithm utilized complex mathematical techniques, specifically Fourier transforms and wavelet analysis, to filter out background noise and isolate faint signals. By clustering the data and analyzing time-based fluctuations, VARnet could detect objects whose brightness changed too subtly or unpredictably for human observers to catch, allowing it to flag anomalies with remarkable precision.[4]

The computational efficiency of the model was equally impressive. Operating on standard graphics processing units (GPUs), the AI achieved sub-millisecond processing times for each individual light curve. In just six weeks, the system flagged 1.9 million variable sources, of which 1.5 million had no prior record in any existing astronomical catalog, immediately distinguishing them as fresh targets for research.[3][4]

The scientific community quickly recognized the rigor of Paz's work. In late 2024, his methodology and findings were published in The Astronomical Journal. Being the sole author of a peer-reviewed paper in a premier astrophysics journal is an exceptionally rare feat for a high school student, validating the robustness of his AI pipeline and his rigorous validation process.[4]

The accolades did not stop at publication. In March 2025, Paz was awarded the $250,000 first-place prize at the Regeneron Science Talent Search, the nation's oldest and most prestigious STEM competition for high school seniors. The award highlighted how his interdisciplinary approach successfully bridged the gap between computer science and observational astronomy.

The breakthrough underscores a broader shift in the scientific landscape: the democratization of artificial intelligence. Tools and computing power that once required massive institutional funding are now accessible enough that a dedicated teenager with a laptop and open-source data can fundamentally advance human knowledge, bypassing traditional academic hierarchies.[3]

Mentorship played a crucial role in this democratization. Kirkpatrick and other Caltech researchers provided the guidance necessary to channel Paz's raw coding ability into rigorous scientific inquiry. "If I see their potential, I want to make sure they reach it," Kirkpatrick noted, emphasizing the importance of taking young researchers seriously and providing them with real-world challenges.[1][3]

Today, Paz is officially employed by Caltech's IPAC while he finishes his high school education. He and Kirkpatrick are preparing to release the complete "VarWISE" catalog to the global astronomical community, which will likely guide future observations by the James Webb Space Telescope and the upcoming Vera C. Rubin Observatory.[1][4]

Beyond the stars, the underlying architecture of Paz's AI model holds immense potential for Earth-bound applications. Because the algorithm is fundamentally designed to detect subtle anomalies in time-series data, researchers believe it could eventually be adapted for use in financial forecasting, environmental monitoring, and even neuroscience, proving that tools built for the cosmos can solve problems much closer to home.[1]