We introduce a fully automated framework that uses LLMs to seamlessly retrieve auxiliary tasks and employs adaptive weighting to integrate them for improved molecular property prediction.