DoorDash uses LLM juries to build food metadata

DoorDash engineers explored new ways to enrich food data. They employed large language models as “jurors” to evaluate entries. The juries assess relevance and accuracy of metadata. Multimodal AI techniques were integrated to handle images and text. Context optimization improves the consistency of the resulting data. The approach aims to scale metadata creation across menus. Results show higher coverage compared with manual tagging. DoorDash plans to refine the system for broader deployment.