Abstract<\/p>\n
In 1998, Amazon.com invented item-to-item collaborative filtering \u2014 the algorithm behind \u201ccustomers who bought this item also bought\u201d \u2014 fundamentally changing how online retailers connect people with products. That innovation replaced earlier user-based approaches with a scalable, item-based similarity model and became one of the most influential recommender systems in the history of e-commerce. In the\u00a0nearly three\u00a0decades since, online retail personalization has evolved through successive waves of machine learning: from matrix factorization and contextual bandits, through deep learning ranking models\u00a0operating\u00a0at massive scale, to today\u2019s frontier \u2014 generative AI systems that automate entire merchandising workflows end to end.\u00a0<\/p>\n
This panel brings together technology and business leaders from Amazon to explore how AI has moved far beyond recommendation engines. Modern e-commerce AI systems\u00a0don\u2019t\u00a0just suggest products \u2014 they generate shopping themes, curate product collections, create marketing copy and imagery, enrich catalog data at scale, and personalize the storefront for hundreds of millions of customers. The panel will trace the technical evolution that enabled each leap \u2014 from collaborative filtering to deep learning to LLMs fine-tuned with domain-specific representations like hierarchical semantic IDs \u2014 while examining the business implications: how AI-driven automation reshapes customer engagement, what new economics emerge when merchandising scales from hundreds of manually curated campaigns to thousands of AI-generated experiences, and how organizations must evolve their strategies and operating models to deploy these systems responsibly at scale.<\/p>\n
<\/p>\n
Target Audience<\/p>\n
This panel targets researchers, applied scientists, and ML engineers interested in recommender systems, personalization, or generative AI who want to understand how Amazon transitions ML prototypes and research ideas into production systems serving hundreds of millions of customers at scale. It will also be valuable for systems engineers and architects interested in the operational challenges of deploying AI-driven automation at\u00a0massive\u00a0scales.<\/p>\n
Graduate students, PhD candidates, and postdoctoral researchers\u00a0in AI\/ML will gain insight into how Amazon\u00a0operates, the types of technical challenges engineers and scientists solve in production environments, and the career paths available across engineering\u00a0and\u00a0applied science roles.<\/p>\n
Attendees should have a working familiarity with core machine learning concepts (supervised learning, collaborative filtering, neural networks) and a general understanding of how recommender systems\u00a0operate\u00a0in production. Prior exposure to reinforcement learning or large language models is helpful but not essential. No programming skills are\u00a0required.<\/p>\n
<\/p>\n
Presenters<\/p>\n
Carmen\u00a0Nestares<\/strong>\u00a0leads merchandising strategy and marketing technology for Amazon\u2019s North America Stores and US Prime. As a VP, she leads engineering, science, UX, product and creative teams. Prior to this role, she served as CMO of Amazon Fashion, CMO of Amazon Home, and general manager for both Amazon Kitchen and the Amazon Wedding Registry program. Previously, she worked at Johnson &\u00a0Johnson\u00a0consumer products, where she spent 12 years leading a variety of personal care and global beauty categories, as well as the acquisition and divesture of several brands.\u00a0Nestares\u00a0began her career in finance as a futures and options broker in Europe. She holds a BS from the University of Granada, Spain, and an MBA from New York University.<\/p>\n Kira Shabalin<\/strong> is a Director of Software Development leading US Prime and Marketing Technology (UPMT) at Amazon. In this role, she oversees engineering, science and product teams building GenAI-powered merchandising solutions that transform how customers discover products on Amazon. Under her leadership, UPMT has pioneered agentic merchandising, which automates the entire merchandising workflow from theme generation and product curation to personalization to campaign delivery. Prior to this role, she served as Director of Amazon Fashion, where she launched over 50 features for visual and personalized product discovery,\u00a0participated\u00a0in establishing Amazon\u2019s Madrid development center, and delivered the “Try Before You Buy” (Prime Wardrobe) program across multiple countries. She also led development of computer vision models that backfill 1.3B+ catalog attributes weekly to improve Amazon\u2019s catalog quality. Previously, she held leadership positions at GoDaddy as Senior Director of Engineering for e-commerce platforms\u00a0operating\u00a0across 56 markets, and at Microsoft as Director of SQL Azure Services Engineering, where she delivered continuous deployment solution for SQL Azure and achieved tens of millions in cost savings. Kira holds an MS in Computer Applications from Technion (Israel Institute of Technology) and an MS in Applied Mathematics from Kazakh State University.\u00a0<\/p>\n Alexandros\u00a0Karatzoglou<\/strong>\u00a0is a Principal Applied Scientist and Senior Science Manager specializing in machine learning and AI-driven personalization. In his role, he leads research and engineering teams focused on advanced recommender systems, reinforcement learning, and large-scale customer modeling. Prior to joining Amazon, he served as a Staff Research Scientist at Google DeepMind, where he advanced deep learning techniques for global recommendation algorithms. Previously, he was the Scientific Director at Telef\u00f3nica Research in Barcelona, leading the lab\u2019s evolution into a pioneer of neural-based collaborative filtering.\u00a0Karatzoglou\u00a0is also a prolific contributor to the open-source community as the author of\u00a0kernlab, a widely used R package for kernel-based machine learning. His academic career includes serving as a professor of Deep Learning at the Barcelona School of Economics. He holds a PhD in Machine Learning from the Vienna University of Technology.<\/p>\n Brent Smith<\/strong> has been engaged in fast-paced, customer-focused product innovation at Amazon for over 26 years. The first 16 years were spent hands-on developing and\u00a0managing in\u00a0Amazon’s Personalization and Recommender Systems org in roles from Software Developer to Director of Personalization. Brent was a co-author of Amazon’s seminal 2003 paper on item-to-item collaborative filtering, which won IEEE Internet Computing’s “Test of Time” award in 2017, and he co-authored the invited “Two Decades of Recommender Systems at Amazon.com” retrospective in the same journal. Recognizing that A\/B testing was an essential part of the innovation cycle, Brent spent over 20 years developing the science and systems for experimentation at Amazon\u2014including leading the\u00a0Weblab\u00a0organization from 2018-2020, which enables tens of thousands of Amazonians to make data-driven product decisions. Today, as a Sr. Principal Applied Scientist, he serves as a company-wide expert helping flagship programs solve complex experimentation and measurement challenges. Having\u00a0witnessed\u00a0the entire evolution from collaborative filtering to modern generative AI systems, Brent brings a unique long-term perspective spanning product strategy, applied science, and engineering at scale.<\/p>\n Ido Rosen<\/strong> is a Sr Principal Applied Scientist in Core AI at Amazon, with prior engineering leadership at Google and experience as a founder of a quantitative trading firm. He combines deep expertise in machine learning, probabilistic programming, computational neuroscience and economics with hands-on skills building low-latency, distributed and HPC systems for production AI and experimentation. His background spans national labs and academia (Argonne, UChicago) through high-frequency trading and cloud-scale AI, giving him rare cross-domain fluency between modeling and infrastructure. A former PhD student who left academia for industry, he maintains an active technical open-source presence focused on ML tooling and PPL-IR. Based in Palo Alto, he brings 14 years of experience translating quantitative science into scalable, latency-sensitive production systems. <\/p>\n Felipe Bertrand<\/strong> is a Senior Engineering Manager at Amazon, leading recommendations, category merchandising, and brand merchandising teams within Amazon’s US Prime and Marketing Technology (UPMT). In this role, he oversees engineering and science teams building AI-powered automated merchandising workflows\u00a0to support marketing events\u00a0at scale. Prior to this role, he led Amazon’s\u00a0EU Item Data Quality organization in Madrid, where his team developed product\u00a0data\u00a0catalog enrichment tools for worldwide markets.\u00a0Before joining Amazon, he\u00a0worked in\u00a0Deimos Space, a company he co-founded\u00a0in 2001, where he managed satellite data processing projects for Global Navigation, Defense and Earth Observation\u00a0missions, including\u00a0European\u00a0Space\u00a0Agency\u00a0missions\u00a0Galileo, Sentinel-3, and Metop-SG. He holds a PhD in Computer Science from Indiana University and an MS in Telecommunication Engineering from the Universidad Polit\u00e9cnica de Madrid.<\/p>\n <\/p>\n","protected":false},"excerpt":{"rendered":" Abstract In 1998, Amazon.com invented item-to-item collaborative filtering \u2014 the algorithm behind \u201ccustomers who bought this item also bought\u201d \u2014 fundamentally changing how online retailers connect people with products. That innovation replaced earlier user-based approaches with a scalable, item-based similarity model and became one of the most influential recommender systems in the history of e-commerce….<\/p>\n","protected":false},"author":2627,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-444","page","type-page","status-publish","hentry"],"yoast_head":"\n