Building a Data Foundation for Multimodal Foundation Models

In this 34-minute conference talk from Data Council, Ethan Rosenthal from Runway's Technical Staff shares valuable insights on constructing a petabyte-scale multimodal feature lakehouse. Explore the unique challenges and requirements of multimodal data curation that extend beyond conventional machine learning systems. Gain practical perspectives on striking the right balance between data quality and quantity while building systems that effectively support both analytical querying and high-performance feature serving for foundation model training. This presentation is particularly valuable for engineers managing heterogeneous data or supporting large-scale distributed training jobs. The talk is part of Data Council's 2025 Day 2 Foundation Models series, where industry experts share knowledge about cutting-edge data and AI systems.