Dynamic Moments: Weaving LLMs into Deep Personalization at DoorDash

QCon San Francisco 2025

Session

Dynamic Moments: Weaving LLMs into Deep Personalization at DoorDash

Tuesday Nov 18 / 03:55PM PST, Ballroom BC at Hyatt Regency, San Francisco

Abstract

In this talk, we’ll walk through how DoorDash is redefining personalization by tightly integrating cutting-edge large language models (LLMs) with deep learning architectures such as Two-Tower Embeddings (TTE) and Multi-Task Multi-Label (MTML) models.

We'll begin by sharing how delivering truly dynamic, moment-aware experiences first requires a robust, structured understanding of our product catalog. We'll detail how we automated the generation of a product knowledge graph using LLMs, enhanced with fine-tuning and agentic workflows.

From there, we'll demonstrate how the synergy between LLMs and DL-powered personalization unlocks new frontiers in dynamic content generation — powering collections, modules, and real-time app assets designed to win users’ “moments” (e.g., movie nights, flu seasons, study snacks). Our framework turns DoorDash into a moment-ready utility that understands and adapts to users’ evolving needs — becoming a one-stop shop for whatever life throws at them.

Key Takeaways:

  1. A practical framework for integrating LLMs with existing deep learning systems like TTE and MTML for real-time personalization.
  2. A blueprint for building and maintaining automated product knowledge graphs using fine-tuned LLMs and agentic pipelines.
  3. How to personalize app experiences dynamically to win high-intent user moments and occasions.

Interview

The majority of our time is spent on designing and scaling systems that combine deep learning-based personalization with generative AI to create adaptive, real-time user experiences. This includes building LLM-powered agents for product enrichment, optimizing MTML/TTE pipelines, and experimenting with end-to-end personalization loops across content, UX, and ranking.

At DoorDash, we’ve seen first-hand how marrying structured predictive models with the generative capabilities of LLMs unlocks entirely new capabilities in product personalization and customer engagement. We believe this is a direction many in the ML community are just beginning to explore, and we want to share our lessons — both technical and organizational — to help others accelerate this convergence. QCon offers the ideal audience for these learnings: technically deep, practically oriented, and curious about real-world AI deployments at scale.

76% senior dev or higher
1:11 speaker ratio
60+ practitioners

QCon San Francisco 2025 is a three day conference for senior software engineers, architects and team leads. An international program committee of working engineers selects every session. Patterns and practices, not products and pitches.

Share

From the same track

Tuesday 18 November

10:35 Ballroom BC Session AI Agents Engineering at AI Speed: Lessons from the First Agentically Accelerated Software Project Adam Wolff Engineer and Individual Contributor to Claude Code @Anthropic, Previously @Robinhood, @Facebook 11:45 Ballroom BC Session Engineering AI for Creativity and Curiosity on Mobile Bhavuk Jain Tech Lead @Google 13:35 Ballroom BC Session AI/ML Improving Meta Generative Ad Text using Reinforcement Learning Alex Nikulkov Research Scientist (RL lead for Monetization GenAI) @Meta 14:45 Seacliff ABC Session AI Agents From Content to Agents: Scaling LLM Post-Training Through Real-World Applications and Simulation Faye Zhang, Andi Partovi 15:55 Ballroom BC Session Dynamic Moments: Weaving LLMs into Deep Personalization at DoorDash Sudeep Das, Pradeep Muthukrishnan 17:05 Ballroom BC Session AI/ML Automating the Web With MCP: Infra That Doesn’t Break Paul Klein Founder @Browserbase, previously Director of Self-Service & Engineering Manager @Mux, Co-Founder & CTO @Stream Club, Technical Lead @Twilio Inc.