Sedentary behavior, obesity, and metabolic syndrome represent a significant global public health challenge. This presentation explores how personalized nudges, delivered as pop-up notifications, offer a low-friction, effective strategy to promote increased physical activity. Maintaining user engagement requires careful consideration of content, timing, and frequency personalization. Current health apps often struggle with sustained user interest due to their static nature and lack of content adaptation. PEARL, leveraging Large Language Models (LLMs) and Reinforcement Learning (RL), dynamically adjusts content and nudge timing based on individual walking patterns. Our large-scale longitudinal user studies demonstrate that RL achieves a threefold improvement in effectiveness compared to rule-based and random nudge selection strategies commonly employed in existing health apps. The presentation will also provide an overview of key research areas being pursued by Google Deepmind India, highlighting recent projects.
Narayan is a researcher at Google since 2010 working on multiple Google products & research topics. Narayan works across broad health AI topics ranging from medical imaging, digital health and LLMs for fitness coaching. Lately, he is working on LLM agent quality and understanding. Previously, Narayan graduated from IISc specializing in ML and Multicore compiler optimization.