Uploaded September 2026 | Updated November 2026, 8 hours ago
In agentic post-training, reward bugs become model behavior.
At Ray Summit 2026, Radhika Gaonkar, Applied Researcher at Prime Intellect, presents a Ray-based system for stress-testing reward designs and model checkpoints before scaling RL runs: distributed rollout sweeps, paired evaluations, failure traces, and CI-style gates that catch proxy hacks before training scales. Code: github.com/RGaonkar/ray-agent-reward-lab
You'll leave with a working approach to validating rewards before they cost you a training run.
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
In agentic post-training, reward bugs become model behavior.
At Ray Summit 2026, Radhika Gaonkar, Applied Researcher at Prime Intellect, presents a Ray-based system for stress-testing reward designs and model checkpoints before scaling RL runs: distributed rollout sweeps, paired evaluations, failure traces, and CI-style gates that catch proxy hacks before training scales. Code: github.com/RGaonkar/ray-agent-reward-lab
You'll leave with a working approach to validating rewards before they cost you a training run.
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










