TRACE: Agentic Catalog Enrichment with Multi-source Evidence Grounding
TRACE, a novel framework for automated catalog attribute enrichment using agentic Large Language Models (LLMs), triangulates multimodal evidence across merchant catalogs, syndicated feeds, and identity-matched web search to propose candidate attribute values with supporting evidence, and verifies the proposed value aga...