In May 2025 I represented Team Canada at the 75th Regeneron International Science and Engineering Fair in Columbus, Ohio, the largest pre-college science competition in the world, with nearly 1,700 finalists from more than 60 countries across 22 categories. I presented NEPTUNE: N-body Exoplanet Prediction Using TTV for Unseen Exoplanets, and it received a Third Award in Physics and Astronomy.

NEPTUNE looks for planets that never pass in front of their star. Most known exoplanets were found by the transit method, which only sees worlds whose orbits happen to line up edge-on with our view, so the catalogue is full of systems that appear to hold a single planet and almost certainly do not. An unseen outer planet still makes itself known: it tugs on the inner transiting planet, and its transits start arriving early or late. Fitting that timing pattern with N-body simulations, machine-learned priors, and Bayesian inference recovers the hidden planet's mass, period, and eccentricity. I've written up the full method and results separately, including the blind test where it recovered the known hidden planet Kepler-46c to within 99.7% on period, in NEPTUNE: Finding Unseen Exoplanets from Transit-Timing Variations.

The project took about 800 hours across a year, on weekends and weekday evenings.

The poster did not fit

ISEF supplies standard 48x48-inch boards. Mine ran taller than that, so I stacked two trifold boards on top of one another to get everything up. Every project then clears two display-and-safety inspectors before it can stay on the floor.

The written feedback afterwards said the poster had too much on it: be more selective, and add simpler figures at the start for readers who are not astronomers. Both notes were fair. The board did not physically fit for the same reason the content did not fit a reader. I had tried to put the entire project on it.

Seven judges, fifteen minutes each

Judging ran as two roughly three-hour blocks. I was handed a schedule of assigned judges and time slots, though unannounced and special-award judges could appear at any point, so there was no part of the day when I was not potentially being assessed.

No two of the seven worked the same way.

The first skipped the presentation entirely and opened with questions, pointing at parts of the poster and asking me to explain them: walk me through the Bayesian inference, explain random forest. He showed no reaction to anything I said, which is its own kind of pressure.

The second and third let me present, interrupting only to clarify.

The fourth stopped me midway to press on a single result: how do I know Kepler-1710 is in resonance, and how can I be sure? He also told me plainly that he did not have the subject background and would have to take my word for parts of it. His own background was in open science for students, so I leaned into the open-source side of the project and the training modules I had written. He did not take the hook.

The fifth had read the board and my supplementary materials the night before and asked for a four-minute summary. The sixth arrived with questions written down from the previous day, on Lomb-Scargle, orbital drift, and unexpected results. The seventh let me get about seven minutes in, then stopped me every sentence or two to define a term or push one step further: how does Lomb-Scargle actually work, how does random forest regression work, how was the Bayesian setup built.

I had prepared hardest for questions on binning, the physics of the N-body simulations, my parameter ranges, and Nyquist sampling. Not one of those came up.

The questions I could not fully answer

The most useful thing I brought home was a list of things I answered badly.

One judge pointed out that in a two-body system orbits precess anyway, once you account for general relativity, a stellar quadrupole moment from the star not being a perfect sphere, or tidal forces. NEPTUNE reads apsidal precession as evidence of an unseen perturbing planet. How much of that drift could come from physics that has nothing to do with a second planet? I did not have a complete answer. It is the sharpest challenge the method faces, and I am still working on it.

Others asked whether 80,000 simulations risks overfitting, and whether the solution simply sits somewhere inside the training set. And whether a model trained on synthetic data with Gaussian noise added can be trusted on real observations. Those are fair questions and they are not yet closed.

One asked why I did not have a principal investigator.

What was actually new

The written feedback singled out the part I had been least certain would land: "Discussion on degeneracies using synodic period and synodic amplitudes is novel and innovative." That is the piece of NEPTUNE I would defend hardest, and an expert reader identified it without being pointed at it.

Judged against quantum computing

Physics and Astronomy had close to 80 projects, spanning quantum computing, artificial intelligence, spectroscopy, and astrophysics. The judges are physicists, but most of them do not work on exoplanets, so before I could explain how NEPTUNE worked I had to explain why transit-timing variations are worth caring about at all.

The following day was public outreach, when families from Columbus and beyond walked the floor. The same poster had to work for seven specialists and for a ten-year-old who had never heard of a transit. Those are genuinely different problems, and doing both in two days taught me more about explaining my work than any talk I had given before.

Team Canada

Eight of us were selected to represent Canada, travelling from British Columbia, Ontario, and Saskatchewan, with two delegates accompanying us. Team Canada came home with three Grand Awards: Tanvir Mundra in Earth and Environmental Sciences, Vincent and Veronica Guo in Environmental Engineering, and NEPTUNE in Physics and Astronomy.

The science was finished before I got on the plane, and a week in Columbus did not improve it. What the week tested was something else, which is whether I could defend it, cold, to people with no stake in whether it was any good. That is a different skill from building it, and it is the one I came home with.

Read the method: NEPTUNE: Finding Unseen Exoplanets from Transit-Timing Variations Project page: PHYS055 on isef.net Team Canada results: Youth Science Canada