Data pipeline at Petabyte scale: Production Video Curation with Ray on CoreWeave

Fragmented multimodal training causes high latency and low GPU utilization. Anyscale and CoreWeave will present a unified architecture that eliminates these silos. By running Anyscale's managed Ray natively on CoreWeave Kubernetes Service (CKS) with CoreWeave AI Object Storage (CAIOS), teams can streamline video curation at petabyte scale. We will share benchmarks showing gains in throughput, utilization, and TCO, providing a reusable reference architecture for efficient, integrated training.