The project proposes an in-depth exploration of advanced machine learning techniques for the automated classification of stellar spectra, aiming to significantly reduce computation time compared to current methods used by Astro+. Beyond improving efficiency, the project envisions the development of models capable of directly determining stellar parameters through deep learning, eliminating the need for manual model fitting. This approach would represent a major step toward fully automated and more efficient spectroscopic analysis.

The initiative is designed to scale alongside forthcoming large-scale spectroscopic surveys such as WEAVE, 4MOST, and MOONS, which will collectively produce millions of high-quality stellar spectra. By progressively adapting models to the growing datasets, the project ensures readiness to process new data as it becomes available. Ultimately, with massive training datasets and ongoing computational advances, the research anticipates the creation of neural networks capable of tackling more complex tasks—such as the automatic detection of binary systems and the unsupervised discovery of stellar properties beyond the scope of traditional techniques.