Uploaded November 2025 | Updated September 2026, 1 week ago
At Ray Summit 2025, Yi Sheng Ong and Eric Higgins from Applied Intuition share how the company scales massive inference and reinforcement learning workloads operating on petabytes of autonomous driving sensor data.
They begin by outlining Ray’s role within Applied’s ML infrastructure, highlighting how it enables unified, distributed execution across Kubernetes clusters. They then dive into how Ray Data powers large-scale batch inference pipelines—streaming raw sensor data from their lake, executing CPU-intensive transformations, and seamlessly feeding the results into GPU inference at scale.
Next, they explore how Ray’s distributed execution model and RLlib support scalable open- and closed-loop reinforcement learning. This includes running thousands of parallel rollouts, colocating GPU learners with simulators for maximum efficiency, and restoring full training state with minimal overhead.
The talk also covers Applied Intuition’s real-world experience managing Ray in production, offering practical guidance for teams applying Ray to large-scale inference and RL workloads.
Liked this video? Check out other Ray Summit breakout session recordings youtube.com/playlist?list=PLzTswPQNepXllnU0C36WtkC0dqkAoDulh
Subscribe to our YouTube channel to stay up-to-date on the future of AI! youtube.com/c/anyscale
🔗 Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com
At Ray Summit 2025, Yi Sheng Ong and Eric Higgins from Applied Intuition share how the company scales massive inference and reinforcement learning workloads operating on petabytes of autonomous driving sensor data.
They begin by outlining Ray’s role within Applied’s ML infrastructure, highlighting how it enables unified, distributed execution across Kubernetes clusters. They then dive into how Ray Data powers large-scale batch inference pipelines—streaming raw sensor data from their lake, executing CPU-intensive transformations, and seamlessly feeding the results into GPU inference at scale.
Next, they explore how Ray’s distributed execution model and RLlib support scalable open- and closed-loop reinforcement learning. This includes running thousands of parallel rollouts, colocating GPU learners with simulators for maximum efficiency, and restoring full training state with minimal overhead.
The talk also covers Applied Intuition’s real-world experience managing Ray in production, offering practical guidance for teams applying Ray to large-scale inference and RL workloads.
Liked this video? Check out other Ray Summit breakout session recordings youtube.com/playlist?list=PLzTswPQNepXllnU0C36WtkC0dqkAoDulh
Subscribe to our YouTube channel to stay up-to-date on the future of AI! youtube.com/c/anyscale
🔗 Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com
