I spend most of my time on AI these days as it seems to have the perfect blend of hard problems to solve combined with the impact it can make to the world.
Before this, I spent years in the trenches of distributed systems — fault tolerance at scale, consensus protocols, virtual clocks, and the various ways to reason about consistency. Tons of fun (and hard) problems that are well understood by the community at this point, and luckily somewhat applicable to AI.
Note I also go by Ryan Ng.
Feel free to support my baby 𝕏 handle
Macrohard (Jan 2026) I improved the reasoning data and trained the foundation model for Macrohard, xAI's initiative to build a human emulator and a software company run entirely by AI.
Grok 4 Fast (Sept 2025) I advocated for, and was part of the tiger team that shipped Grok 4 Fast, bringing orders-of-magnitude improvements in cost and reasoning efficiency.
Grok 4 (Jul 2025) I improved and shipped parallel test-time compute for Grok 4, the flagship reasoning model.
Sora (Jan 2024) Shipped Sora 1 as the #1 core contributor, starting as a forward-deployed engineer before transitioning to the project full-time.
Ray (Mar 2023) Joined as a TL for Anyscale / Ray. Developed the algorithm for optimistic memory scheduling for ML workloads.
Style transfer (Jun 2022) I built AI prototypes that leverage GANs to personalize content at scale.
Portal / smart camera (Jun 2020) TL for Augmented Reality technology and frameworks at Building 8, Facebook's DARPA-inspired hardware lab founded by Regina Dugan. Shipped the smart camera and AR experiences on Portal, the flagship video product.
Kubernetes (Jul 2015) Core contributor for Google's cluster management systems, transitioning the team from Borg to Kubernetes.