Recommendation and search systems are now used in high-stakes settings, including to help find jobs, schools, and partners. Building public interest recommender systems in such settings bring both individual-level (enabling exploration, diversity, data quality) and societal (fairness, capacity constraints, algorithmic monoculture) challenges. In this talk, I'll discuss our theoretical, empirical, and deployment work in tackling these challenges, including ongoing work on (a) applicant behavior and recommendations for the NYC HS match, (b) a platform to help discharge patients to long-term care facilities, (c) feed ranking algorithms on Bluesky for research paper recommendations. Finally, I’ll discuss our efforts to use sparse autoencoders to derive natural language concepts that can form the foundation of interpretable, explorable, and steerable recommender and search systems.

Speaker Bio: Nikhil Garg is an Assistant Professor of Operations Research and Information Engineering at Cornell Tech as part of the Jacobs Institute. He uses algorithms, data science, and mechanism design approaches to study democracy, markets, and societal systems at large. Nikhil has received the NSF CAREER, INFORMS George Dantzig Dissertation Award, an honorable mention for the ACM SIGecom dissertation award, several other best paper awards, and Forbes 30 under 30 for Science. He received his PhD from Stanford University and has spent considerable collaborating with government agencies and non-profits.

Grounded in psychology and user research from Netflix’s first test of a conversational recommendation system, this session invites participants to step outside the technical box and consider personalization as a felt experience. We’ll explore the often-unseen but deeply revealing ways people express intent, interpret results, and find value in participating in algorithmically powered discovery. Drawing on years of collaboration with ML and AI teams, this talk offers nuanced perspectives and practical takeaways for building recommendation systems that meet users where they are to co-create more human-centered search and discovery experiences.

Speaker Bio: Kelley Robinson is a user experience researcher and strategist specializing in personalization, AI-driven experiences, and complex product ecosystems. A futurist at heart, pathologically curious, and optimistically invested in technology’s potential to serve each person and their unique needs, she delights in generating insights from quantitative data, qualitative research, and what lives between the lines. Her work bridges design, engineering, and data science to inspire innovation and help product teams make more human-centered decisions. Kelley holds a PhD in Social and Personality Psychology and completed a postdoctoral fellowship studying the science of social connection across romantic, platonic, and caregiving relationships. She began her tech career at Facebook (Meta), where she worked on public content and video experiences, including the launch and development of Facebook Live and real-time interaction between broadcasters and viewers. She is currently a Senior Researcher at Netflix, where she leads consumer insight efforts that shape algorithmically driven content discovery experiences, including personalized recommendations, explanatory evidence, and generative search.

Large recommender systems power the world's biggest online media platforms, finding patterns in past engagement data in order to match billions of viewers to billions of pieces of content. This, however, requires effective exploration of an ever-increasing catalogue of items, and this talk explores some of our strategies and how they are applied in production settings. In this talk, I would dive deeper into 1) Uncertainty estimations and their applications in contextual bandit settings, 2) Item-centric exploration systems and 3) A set of metrics to measure RL based recommender systems and evaluate the impact of exploration against these metrics.

Speaker Bio: Arnab (Arnie) Bhadury (arniebh@google.com) is a Staff Machine Learning Engineer at YouTube Shorts and has been in the organizing committee of ACM RecSys for the past four years. Prior to YouTube, he worked on fitness recommendation at Peloton and news recommendation at Flipboard. His interests lie in large scale recommendations, content understanding and Bayesian machine learning. Arnie has been an active member of the recommender systems community, and has been an active PC member at RecSys, KDD, SIGIR and WWW. Arnie has also hosted several machine learning workshops and events in his home city of Vancouver, British Columbia, Canada

Effective product discovery is crucial for user satisfaction and business growth in diverse marketplaces like DoorDash, which span grocery, retail, and other merchandise verticals. We present a personalization framework aimed at enhancing discovery by dynamically balancing three user value dimensions: price sensitivity (affordability), established tastes (familiarity), and the desire for new items (novelty). Our system combines efficient dense item retrieval with a learned multi-objective ranking model that optimizes for a combination of conversion, basket size, and exploration metrics. To tackle the challenges of personalizing for users with limited activity, particularly new and low-engagement individuals, we introduce an innovative personalization layer powered by a Large Language Model (LLM). This LLM analyzes rich, cross-vertical behavioral signals, including restaurant order history, to infer high-intent shopping missions and proactively surface relevant category collections. This approach enables meaningful personalization even when traditional item-level history is sparse. We will discuss the practicalities of scaling this framework in a production setting and share key learnings from our evaluations, illustrating how explicitly addressing utility trade-offs through multi-objective optimization and LLM-driven inference leads to significantly improved, diverse, and engaging discovery outcomes.

Speaker Bio: Raghav Saboo is a Staff Machine Learning Engineer at DoorDash, leading Personalization related ML applications within the New Verticals business line. He was previously at Amazon, where his main focus was on developing the first set of large language models and distilled models for new language launches for Alexa AI. Prior to that Raghav worked as a Machine Learning consultant building multiple zero to one solutions for clients across industries. He holds a Masters from Duke University and a combined BEng and MEng from Imperial College London.

Sudeep Das is the Head of Machine Learning, New Business Verticals, at DoorDash, leading Consumer ML, and Product Knowledge Graph ML, and Fulfillment and Inventory ML teams within the New Verticals. He was previously a Machine Learning Leader at Netflix, where his main focus was on developing the next generation of machine learning algorithms to drive the personalization, discovery and search experience in the product. Sudeep has had more than twenty years of experience in machine learning applied to both large scale scientific problems, as well as in the industry. He is a frequent speaker at RecSys, SIGIR, ICML, ReWork, MLConf, Nordic Media Conference, and other machine learning conferences. He holds a PhD in Astrophysics from Princeton University.