Nixon Dinh

As search shifts from classic keyword matching to more conversational experiences, product data quality becomes critical to LLM-powered retrieval. At PayPal, we tested how enriching traditional catalog data could help AI systems better find, understand, and rank products across large-scale commerce catalogs. We built a RAG-based AI judge to compare enrichment approaches and identify five patterns that consistently improved AI discovery results.In this talk, we'll share the evaluation framework, key lessons, and a practical approach for preparing enterprise data for conversational and agentic search.