Why DoorDash, Instacart, and Uber Eats Integrated LLMs Into Search Three Different Ways

DoorDash, Instacart, and Uber Eats each integrated LLMs into their food delivery search systems differently, revealing how existing infrastructure shapes architectural choices more than model selection does. DoorDash uses LLMs mostly offline to enrich a knowledge graph, keeping runtime retrieval classical. Instacart consolidates multiple specialized models under a unified LLM strategy, splitting head queries to an offline RAG-and-cache pipeline and tail queries to a real-time fine-tuned Llama-3-8B. Uber Eats goes deepest, fine-tuning a Qwen LLM as the embedding backbone of a two-tower retrieval system covering all verticals and languages. Key universal tradeoffs include hybrid classical+LLM systems, domain-specific fine-tuning or RAG on top of general world knowledge, and guardrails to keep outputs catalog-aligned.