Distill the LLM, Don't Serve It: Search & Personalization at DoorDash — Raghav Saboo, DoorDash

Why watch

Raghav Saboo presents DoorDash's search and personalization architecture that uses LLMs as an offline reasoning layer, then distills insights into fast models for serving. Covers four primitives: LLM-generated relevance labels (+2.3% NDCG), semantic IDs as a learned catalog taxonomy (+4-5% MRR), consumer memory with text/vectors/graph, and steerable personalized collections (+1% order rate in pets). For ML engineers building production search and recommendation systems at scale.