Kepler: A Deep Learning Pipeline for Transit-Based Exoplanet Candidate Identification in TESS Sectors 1 through 97

Abstract

TESS has produced light curves for hundreds of millions of stars, more than traditional tools like Box Least Squares can feasibly vet alone. We present Kepler, a CNNtransformer hybrid trained to detect the periodic brightness dips caused by transiting planets. Using TESS Sectors 1 through 97 and a training set of confirmed exoplanets, TESS Objects of Interest, and simulated transit injections, Kepler identifies 14 exoplanet candidates, each with a derived orbital period, transit depth, and signal-to-noise ratio. The project also includes a general-purpose conversational assistant built on a large language model, tuned for strength in physics but able to help with a wide range of topics, giving users a way to explore the candidate data and ask questions alongside the detection results.