DoorDash Uses LLM Jury System to Generate Food Item Metadata at Scale

DoorDash has published a technical blog post detailing how it uses a multi-model large language model approach, dubbed an 'LLM jury,' to build food metadata for its platform. The system leverages multimodal AI and context optimization to extract and generate structured information about menu items. By having multiple LLMs evaluate and cross-check outputs, DoorDash aims to improve accuracy and reliability of food data. This metadata powers features such as search, recommendations, and dietary filtering across the DoorDash app.