TLDR Tech / Marketing
Why DoorDash, Instacart, and Uber Eats Integrated LLMs Into Search Three Different Ways
Food delivery search breaks classical keyword matching in every way at once, from synonyms and typos to compound constraints like "vegan chicken sandwich", which is why DoorDash, Instacart, and Uber Eats each rebuilt it around LLMs differently. DoorDash uses an LLM to enrich an existing knowledge graph offline, then constrains query parsing to only the labels that graph already knows, lifting dish carousel triggers by roughly 30%. Instacart fine-tuned a Llama-3-8B model to handle rare tail queries in real time under 300ms, cutting scroll depth 6% and tail-query complaints in half. Uber Eats went furthest, embedding every query and document through a fine-tuned Qwen model shared across every vertical, market, and language.
Read the full story