Episode 403 - Neural Networks for Real-Time Personalization: Insights with Vedant Agarwal

Real-time personalization is transforming industries, especially e-commerce, where engaging customers and driving conversions are pivotal. This session delves into the innovative use of a two-tier neural ranking architecture comprising L1 and L2 rankers. The L1 stage employs embedding-based indexing to map user behavior and item attributes into a unified semantic space, enabling rapid filtering of potential matches. This broad-level filtering sets the foundation for precision personalization by capturing nuanced relationships within extensive product catalogs.

At the L2 stage, advanced techniques such as sequence modeling and attention mechanisms leverage granular user-specific features, including session context and historical interactions, to deliver refined, adaptive recommendations. This hierarchical approach has been a game-changer for businesses, driving significant improvements in recommendation accuracy, response times, and sales conversions.

This talk will uncover the core challenges and strategies for deploying such architectures in real-time settings, addressing critical components like data ingestion pipelines, microservices-based orchestration, and maintaining sub-50ms latency at scale. Through compelling case studies, attendees will learn how leading companies have successfully integrated L1-L2 rankers to optimize user experiences and achieve measurable business outcomes. From feature engineering to scalable infrastructure design, this session provides actionable insights for organizations looking to excel in personalization technology.