Abstract: As DoorDash expands beyond restaurants into groceries, convenience, and retail, the need for intelligent, scalable solutions has never been greater. Large Language Models (LLMs) are transforming how we personalize experiences, optimize search, and structure product catalogs to create a seamless shopping journey.
In this session, we’ll explore how DoorDash leverages LLMs alongside traditional machine learning techniques to enhance product discovery, improve taxonomy, and streamline order fulfillment. We’ll discuss how LLMs enable rapid basket building, intuitive product substitutions for out-of-stock items, and more accurate search results tailored to individual preferences.
A key challenge in building a high-quality marketplace is ensuring a structured and enriched product catalog. LLMs play a critical role in refining product knowledge graphs, linking related products, and handling ambiguous or incomplete data. By integrating LLMs with existing ML systems, we enhance taxonomy management and create a more robust foundation for search, recommendations, and fulfillment.
Beyond the technical aspects, we’ll share key lessons learned, real-world challenges, and best practices for applying LLMs in dynamic, fast-moving environments. This includes managing data inconsistencies, balancing accuracy with scalability, and ensuring AI-driven decisions align with business goals and customer expectations.
Attendees will gain actionable insights into how LLMs can drive personalization, improve operational efficiency, and power the next generation of e-commerce experiences. Whether you’re an engineer, data scientist, or product leader, this session will provide a practical framework for integrating LLMs into large-scale, real-world applications.
Bio: Bio Coming Soon!
Sudeep Das
Title
Head of Ai, New Business Verticals | DoorDash Inc


