Uploaded September 2026 | Updated November 2026, 3 days ago
Generating millions of synthetic preference pairs for training requires orchestrating GPU inference, rate-limited search APIs, and LLM judges, each with different concurrency profiles and cost constraints.
At Ray Summit 2026, Eric Wang and Gaurav Arora from Pinterest share a composable operator framework on Ray Data where ML teams configure what runs, models, tools, and prompts, while the platform handles how it scales: streaming execution and backpressure across heterogeneous stages. The trained model then becomes a producer for the next round, driving iterative model self-improvement.
You'll leave with production results, architecture decisions, and the quality gains from closing the loop.
Liked this video? Check out other Ray Summit breakout session recordings: youtube.com/playlist?list=PLZNaJyYZRPdo
Subscribe to our YouTube channel to stay up-to-date on the future of AI: youtube.com/c/anyscale
đź”— Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com
Generating millions of synthetic preference pairs for training requires orchestrating GPU inference, rate-limited search APIs, and LLM judges, each with different concurrency profiles and cost constraints.
At Ray Summit 2026, Eric Wang and Gaurav Arora from Pinterest share a composable operator framework on Ray Data where ML teams configure what runs, models, tools, and prompts, while the platform handles how it scales: streaming execution and backpressure across heterogeneous stages. The trained model then becomes a producer for the next round, driving iterative model self-improvement.
You'll leave with production results, architecture decisions, and the quality gains from closing the loop.
Liked this video? Check out other Ray Summit breakout session recordings: youtube.com/playlist?list=PLZNaJyYZRPdo
Subscribe to our YouTube channel to stay up-to-date on the future of AI: youtube.com/c/anyscale
đź”— Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com










