Real-time Feature Generation at Lyft

Explore real-time feature generation at scale through this 58-minute MLOps podcast episode featuring Rakesh Kumar, Senior Staff Software Engineer at Lyft. Discover how Lyft processes tens of millions of events per minute to generate features that accurately reflect current marketplace conditions for optimal operational efficiency. Learn about the evolution from humble beginnings and naive pipelines to sophisticated infrastructure through iterative improvements and strategic trade-offs to achieve low-latency, real-time feature delivery. Examine the critical role of MLOps in managing the lifecycle of real-time feature pipelines, including monitoring and deployment strategies. Understand the practicalities of building and maintaining high-throughput, low-latency systems that power dynamic marketplace and business-critical products. Gain insights into latency optimization tricks, config management complexity, data contract implementation, feature store architecture, offline versus online workflows, decision-making processes during technology shifts, cost evaluation strategies, model feature discussions, hot shard management techniques, and pipeline feature bundling approaches that enable real-time machine learning at enterprise scale.