--- title: How DoorDash Built an AI Shopping Assistant That Doesn’t... description: DoorDash published a three-part engineering deep dive on Ask DoorDash, its conversational AI shopping assistant. The system combines LLMs, specialized domain... canonical: https://daily.dev/posts/how-doordash-built-an-ai-shopping-assistant-that-doesn-t-rely-on-the-llm-alone-w3gyrfrce twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: How DoorDash Built an AI Shopping Assistant That Doesn’t Rely on the LLM Alone | daily.dev og:description: DoorDash published a three-part engineering deep dive on Ask DoorDash, its conversational AI shopping assistant. The system combines LLMs, specialized domain... og:url: https://daily.dev/posts/how-doordash-built-an-ai-shopping-assistant-that-doesn-t-rely-on-the-llm-alone-w3gyrfrce og:image: https://api.daily.dev/og/posts/w3gyrfrce.png og:image:alt: How DoorDash Built an AI Shopping Assistant That Doesn’t Rely on the LLM Alone og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # How DoorDash Built an AI Shopping Assistant That Doesn’t Rely on the LLM Alone **[InfoQ](https://daily.dev/sources/infoq)** · 3 min read · 5 upvotes · 1 comments ## Summary DoorDash published a three-part engineering deep dive on Ask DoorDash, its conversational AI shopping assistant. The system combines LLMs, specialized domain agents, Model Context Protocol (MCP)-based tooling, and a three-tier memory architecture (long-term, session, and agentic memory) to deliver personalized grocery and restaurant discovery experiences. An intelligence layer retrieves relevant memories via semantic vector search and injects them into prompts, separating memory management from model inference. Business logic is exposed through reusable MCP tools backed by existing DoorDash services rather than embedded in prompts. An automated evaluation framework simulates stateful conversations at scale, running over 2,000 daily evaluations and cutting regression testing from six hours to twenty minutes. Early production results include 24% higher grocery checkout conversion, 17% larger basket sizes, 7% fewer conversational turns, and a 35% latency reduction validated without quality regression. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does DoorDash's Ask DoorDash assistant use memory to personalize shopping recommendations? It uses three memory systems: long-term memory generated offline from historical consumer behavior capturing preferences like favorite cuisines and dietary restrictions, session memory maintaining context during a conversation, and agentic memory storing explicit facts users provide. Relevant memories are retrieved via semantic vector search, ranked, and inserted into prompts separately from model inference. _Teams designing memory-backed AI agents can follow architecture breakdowns like this on daily.dev._ ### What results did DoorDash see from adding consumer memory to its AI shopping assistant? Grocery checkout conversion rose by approximately 24%, basket sizes increased by 17%, and conversational turns dropped by 7% during a seven-day evaluation window. Restaurant discovery saw a 15% higher conversion rate on open-ended queries after the memory-backed intelligence layer was introduced. _Engineers weighing AI agent investments can track real production metrics like these via daily.dev._ ### How did DoorDash speed up regression testing for its AI agents? DoorDash built an automated evaluation framework that simulates stateful customer conversations using LLM-generated users and recorded tool fixtures, mirroring the production runtime to test orchestration, guardrails, and domain agents independently. This scaled quality measurement to over 2,000 automated evaluations daily and cut regression testing time from six hours to twenty minutes while improving quality scores by eight points. _Anyone building evaluation pipelines for LLM agents can find case studies like this on daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@kartiknvj** · 0 upvotes > The detail that stands out is the evaluation framework simulating stateful conversations and dropping regression testing from six hours to twenty minutes, because multi-turn eval is where most teams give up and ship on vibes. Keeping business logic in MCP tools backed by real services instead of stuffing it into prompts is also what makes those 2,000 daily evals meaningful, since you are testing the actual system, not a prompt approximation of it. I would be curious how they seed the simulated conversations so the eval set keeps matching real user drift. ## Similar posts on daily.dev - [Building Ask DoorDash \(Part 2\): Intelligence](https://daily.dev/posts/building-ask-doordash-part-2-intelligence-4xcmvaewz) · Doordash · 2 upvotes · 0 comments - [DoorDash Builds LLM Conversation Simulator to Test Customer Support Chatbots at Scale](https://daily.dev/posts/doordash-builds-llm-conversation-simulator-to-test-customer-support-chatbots-at-scale-pijhailvz) · InfoQ · 1 upvotes · 0 comments - [Building a unified consumer memory for personalization at scale](https://daily.dev/posts/building-a-unified-consumer-memory-for-personalization-at-scale-emiywz2l0) · Doordash · 0 upvotes · 0 comments - [Our Early Journey to Transform Instacart’s Discovery Recommendations with LLMs](https://daily.dev/posts/our-early-journey-to-transform-instacart-s-discovery-recommendations-with-llms-wpv1ryxzx) · Instacart · 1 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#mcp](https://daily.dev/tags/mcp), [#rag](https://daily.dev/tags/rag), [#conversational-ai](https://daily.dev/tags/conversational-ai) [View this post on daily.dev](https://daily.dev/posts/how-doordash-built-an-ai-shopping-assistant-that-doesn-t-rely-on-the-llm-alone-w3gyrfrce) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"How DoorDash Built an AI Shopping Assistant That Doesn’t Rely on the LLM Alone","url":"https://daily.dev/posts/how-doordash-built-an-ai-shopping-assistant-that-doesn-t-rely-on-the-llm-alone-w3gyrfrce","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/how-doordash-built-an-ai-shopping-assistant-that-doesn-t-rely-on-the-llm-alone-w3gyrfrce"},"datePublished":"2026-07-13T14:24:18.394Z","dateModified":"2026-09-14T08:44:12.585Z","description":"DoorDash published a three-part engineering deep dive on Ask DoorDash, its conversational AI shopping assistant. The system combines LLMs, specialized domain...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/197007457f343b5b362a9aee87d17d68?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/197007457f343b5b362a9aee87d17d68?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"InfoQ","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"InfoQ","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/afc3bced3e1e4b188dd9127017a60e0c","url":"https://daily.dev/sources/infoq"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/how-doordash-built-an-ai-shopping-assistant-that-doesn-t-rely-on-the-llm-alone-w3gyrfrce","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":5},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"llm,ai-agents,mcp,rag,conversational-ai","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"InfoQ","item":"https://daily.dev/sources/infoq"},{"@type":"ListItem","position":3,"name":"How DoorDash Built an AI Shopping Assistant That Doesn’t Rely on the LLM Alone"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/how-doordash-built-an-ai-shopping-assistant-that-doesn-t-rely-on-the-llm-alone-w3gyrfrce","comment":[{"@type":"Comment","text":"The detail that stands out is the evaluation framework simulating stateful conversations and dropping regression testing from six hours to twenty minutes, because multi-turn eval is where most teams give up and ship on vibes. Keeping business logic in MCP tools backed by real services instead of stuffing it into prompts is also what makes those 2,000 daily evals meaningful, since you are testing the actual system, not a prompt approximation of it. I would be curious how they seed the simulated conversations so the eval set keeps matching real user drift.","datePublished":"2026-07-14T16:30:50.914Z","url":"https://daily.dev/posts/w3gyrfrce#c-VbBaWbz9u","author":{"@type":"Person","name":"kartik-nvjk","url":"https://daily.dev/kartiknvj","image":"https://media.daily.dev/image/upload/s--3gGgsVCw--/f_auto/v1781456774/avatars/avatar_TvTVeiMdkRCqWUDullFmy?_a=BAMAMiWQ0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/how-doordash-built-an-ai-shopping-assistant-that-doesn-t-rely-on-the-llm-alone-w3gyrfrce#faq","mainEntity":[{"@type":"Question","name":"How does DoorDash's Ask DoorDash assistant use memory to personalize shopping recommendations?","acceptedAnswer":{"@type":"Answer","text":"It uses three memory systems: long-term memory generated offline from historical consumer behavior capturing preferences like favorite cuisines and dietary restrictions, session memory maintaining context during a conversation, and agentic memory storing explicit facts users provide. Relevant memories are retrieved via semantic vector search, ranked, and inserted into prompts separately from model inference. Teams designing memory-backed AI agents can follow architecture breakdowns like this on daily.dev."}},{"@type":"Question","name":"What results did DoorDash see from adding consumer memory to its AI shopping assistant?","acceptedAnswer":{"@type":"Answer","text":"Grocery checkout conversion rose by approximately 24%, basket sizes increased by 17%, and conversational turns dropped by 7% during a seven-day evaluation window. Restaurant discovery saw a 15% higher conversion rate on open-ended queries after the memory-backed intelligence layer was introduced. Engineers weighing AI agent investments can track real production metrics like these via daily.dev."}},{"@type":"Question","name":"How did DoorDash speed up regression testing for its AI agents?","acceptedAnswer":{"@type":"Answer","text":"DoorDash built an automated evaluation framework that simulates stateful customer conversations using LLM-generated users and recorded tool fixtures, mirroring the production runtime to test orchestration, guardrails, and domain agents independently. This scaled quality measurement to over 2,000 automated evaluations daily and cut regression testing time from six hours to twenty minutes while improving quality scores by eight points. Anyone building evaluation pipelines for LLM agents can find case studies like this on daily.dev."}}]} ```