Lessons Learned: The Journey to Real-Time Machine Learning at Instacart

This talk details Instacart's transition from batch-oriented machine learning systems to a real-time ML platform. It covers the "why" behind real-time ML, focusing on improving customer experience and business impact. The presentation outlines the key infrastructural changes, including the development of a feature store and an online inference platform, and highlights the implementation of real-time feature pipelines using streaming infrastructure like Kafka and Flink. Several use cases are explored, such as real-time item availability, online query understanding, and contextual recommendations, demonstrating the significant benefits and lessons learned from this journey.