← back
Transforming search and discovery using LLMs — Tejaswi & Vinesh, Instacart
Takeaway
Plug LLMs in on top of engagement-derived candidates rather than letting them classify cold — the seeded prompt aligns LLM commonsense with real user-conversion data.
Summary
- Instacart used LLMs to upgrade their query-understanding stack (10K taxonomy labels, 6K commonly used) because their FastText + NPMI fallback models failed on tail queries lacking engagement data.
- Naive LLM classification gave plausible-sounding but business-wrong results ('protein' → chicken/tofu, not bars/shakes); fix was to feed top-converting categories per query as in-context candidates and let the LLM rerank — precision +18pp, recall +70pp on tail.
- Query rewrites: LLM generates substitute (avocado oil → olive oil), broader (healthy cooking oil), and synonymous rewrites with strong offline gains.
- LLM-generated structured query annotations also power related-item carousels so users no longer dead-end on a single product page.
searchquery-understandingecommerce
Original description
Learn how Instacart uses cutting-edge LLMs to redefine search and product discovery. - Explore innovative solutions overcoming traditional search engine limitations for grocery shopping. - Discover how LLMs enhance user intent understanding and generate engaging content. - See practical applications of LLM technology to improve search relevance and user experience. About Tejaswi Tenneti Tejaswi Tenneti is currently a Director of Machine Learning at Instacart, the north american leader in online grocery. Prior to Instacart, Tejaswi was a tech lead in machine learning teams at Apple and Oracle where he worked on various applications related to Search and Recommendations for local maps data and Enterprise. Tejaswi holds a BS from IIIT, Allahabad and an MS from Stanford University specializing in AI About Vinesh Gudla Vinesh is a Staff Machine Learning Engineer at Instacart on the search and discovery team. He has previously worked on balancing multiple objectives in search in a marketplace and has authored numerous well-received blogposts and articles about his work. He is currently working on bringing Generative AI to production at ecommerce scale at Instacart. Recorded at the AI Engineer World's Fair in San Francisco. Stay up to date on our upcoming events and content by joining our newsletter here: https://www.ai.engineer/newsletter