Learning in Dreams, Winning in Reality: A Continuous Dyna Loop for a MOBA ↗
Training agents in learned world models, with a continuous loop between imagination and real environments.
I’m a research scientist and full-stack ML engineer bridging foundational models, neural simulation, and real-world intelligence.
Currently at Riot Games, I work across the entire machine learning lifecycle—from novel world-model research and evaluation to hyperscale training infrastructure serving billions of predictions. Previously, I built ML systems at Google, Meta, and Uber.
Driven by a deep curiosity for robotics, quantitative systems, and the arts, my work focuses on building expressive, agentic technologies that move seamlessly from theoretical imagination into everyday human experiences.
Research: World models, neural simulation, and reinforcement learning. See my research.
CV: My résumé is available here .
Email: jordy.kieto@gmail.com
Research & engineering at
Training agents in learned world models, with a continuous loop between imagination and real environments.
Compact world models that run interactive driving and flight environments at 30 fps on a laptop CPU.
A multimodal driving policy exploring closed-loop generalization, cross-town transfer, and error recovery.