Accurately mapping user queries to business categories is a fundamental Information Retrieval challenge for multi-category marketplaces, where context-sparse queries such as ''Wildflower'' exhibit intent ambiguity, simultaneously denoting a restaurant chain, a retail product, and a floral item. Traditional classifiers force a winner-takes-all assignment, while general-purpose LLMs hallucinate unavailable inventory. We introduce an Agentic Multi-Source Grounded system that addresses both failure modes by grounding LLM inference in (i) a staged catalog entity retrieval pipeline and (ii) an agentic web-search tool invoked autonomously for cold-start queries. Rather than predicting a single label, the model emits an ordered multi-intent set, resolved by a configurable disambiguation layer that applies deterministic business policies and is designed for extensibility to personalization signals. This decoupled design generalizes across domains, allowing any marketplace to supply its own grounding sources and resolution rules without modifying the core architecture. Evaluated on DoorDash's multi-vertical search platform, the system achieves +10.9pp over the ungrounded LLM baseline and +4.6pp over the legacy production system. On long-tail queries, incremental ablations attribute +8.3pp to catalog grounding, +3.2pp to agentic web search grounding, and +1.5pp to dual-intent disambiguation, yielding 90.7% accuracy (+13.0pp over baseline). The system is deployed in production, serving over 95% of daily search impressions, and establishes a generalizable paradigm for applications requiring foundation models grounded in proprietary context and real-time web knowledge to resolve ambiguous, context-sparse decision problems at scale.

Optimizing industrial search ranking models solely for user engagement signals often introduces systematic biases, prioritizing popular or price-anchored items that may not satisfy semantic intent. We present a production-scale multi-task ranking system that integrates semantic relevance as a primary optimization objective, enabling explicit and controllable relevance--engagement trade-offs. Our architecture employs an ordinal relevance head that predicts cumulative probabilities over relevance thresholds, preserving the inherent ordering of labels. These outputs are integrated with engagement heads through a unified value model scoring function, enabling systematic balancing of semantic quality and short-term behavioral signals. To provide high-quality supervision for this multi-task framework, we utilize fine-tuned lightweight Large Language Models (LLMs) to generate three-level ordinal relevance labels: irrelevant, moderately relevant, and highly relevant. We address challenges regarding label distribution sensitivity and ensure high alignment with human annotations to enable efficient labeling for over 100 million query--item pairs. Evaluation across offline metrics, including NDCG@10, and online A/B experiments demonstrates that our approach significantly improves semantic alignment while preserving core engagement objectives.

Extracting knowledge from unstructured or semi-structured textual information is essential for the machine learning applications that power DoorDash's search experience, and the development and maintenance of its product knowledge graph. Large language models (LLMs) have opened up new possibilities for utilizing their power in these areas, replacing or complementing traditional natural language processing methods. LLMs are also proving to be useful in the label and annotation generation process, which is critical for these use cases. In this talk, we will provide a high-level overview of how we incorporated LLMs for search relevance and product understanding use cases, as well as the key lessons learned and challenges faced during their practical implementation.

In an instant search setting such as Netflix Search where results are returned in response to every keystroke, determining how a partial query maps onto broad classes of relevant entities orfacets --- such as videos, talent, and genres --- can facilitate a better understanding of the underlying objective of that query. Such a query-to-facet mapping system has a multitude of applications. It can help improve the quality of search results, drive meaningful result organization, and can be leveraged to establish trust by being transparent with Netflix members when they search for an entity that is not available on the service. By anticipating the relevant facets with each keystroke entry, the system can also better guide the experience within a search session. When aggregated across queries, the facets can reveal interesting patterns of member interest. A key challenge for building such a system is to judiciously balance lexical similarity with behavioral relevance. In this paper, we present a high level overview of a Query Facet Mapping system that we have developed at Netflix, describe its main components, provide evaluation results with real-world data, and outline several potential applications.

We discuss salient challenges of building a search experience for a streaming media service such as Netflix. We provide an overview of the role of recommendations within the search context to aid content discovery and support searches for unavailable (out-of-catalog) entities. We also stress the importance of keystroke-level Instant Search experience, and the technical challenges associated with implementing it across different devices and languages for a global audience.

At OpenTable, recommendations play a key role in connecting diners with restaurants. The act of recommending a restaurant to a diner relies heavily on aligning everything we know about the restaurant with everything we can infer about the diner. Our methods go beyond using the diner-restaurant interaction history as the sole input -- we use click and search data, the metadata of restaurants, as well as insights gleaned from reviews, together with any contextual information to make meaningful recommendations. In this talk, I will highlight the main aspects of our recommendation stack built with Scala using Apache Spark.