AstroAid

Personalized Target Down-Selection for Amateur Astronomers

Abstract

Selecting observation targets in astronomy requires reasoning over constraints like visibility, brightness, and scientific value. In domains such as variable star monitoring, where thousands of targets exist and time is limited, making informed choices is essential but often overwhelming, particularly for novice amateur astronomers. We present AstroAid, a language model-based assistant to support target down-selection by integrating user preferences, catalog metadata, and observability constraints. The system generates ranked recommendations with natural-language justifications, enabling both autonomous and human-in-the-loop planning. We evaluate AstroAid’s performance on two key dimensions: replicate consistency and persona sensitivity. Results show that AstroAid produces stable, personalized outputs, demonstrating its utility as a decision support tool for constrained observational workflows. While focused on variable star campaigns, this approach generalizes to other sensing contexts where task prioritization, user alignment, and transparent reasoning are essential.