Retail Search: Retail Search with LLM public preview Retail Search with LLM is in public preview

## Feature **Retail Search: Retail Search with LLM public preview** Retail Search with LLM is in public preview. Retail Search with LLM improves ranking by improving AI-driven grading of how relevant each product is for a specific query. Prior to this upgrade, these relevance grades were generated by an older generation of AI that produced imperfect scoring. This sometimes caused low-relevance products to be highly ranked in search results. With this upgrade, Retail Search uses state-of-the-art AI techniques to do the following: * Develop a Giant Relevance LLM that can accurately grade product/query relevance in any retail category in any supported language. * Distill a smaller LLM that is specific to one retailer from the Giant Relevance LLM. This smaller LLM contains the knowledge needed to grade query/product relevance for a specific retailer's catalog and their unique query stream. * Use the smaller, retail-specific LLM to accurately grade products in real-time. * Allow downstream AIs in Retail Search that consume these more accurate query/product relevance grades to correctly rank lower-relevance products as lower in search results. * Lead to increases in revenues and visits for retailers by focusing early search results on higher-relevance products. _Who has access to Retail Search with LLM in the public preview phase?_ You must meet the access criteria to be considered for the Public Preview. You need to have: * A fully onboarded Retail project. This means the project is: * Fully onboarded onto Retail Search * Has stable usage. It is being used for production, has no sudden ramps up/down, and has no off-label (unsupported) usage * Correctly onboarded with no major data quality issues * Sufficient search volumes: * Have >5M searches/day (counting search only, no browse) served by Retail Search for each of the past 30 days If you have multiple projects, you can choose only one project to use for the public preview.