Marco Berlot

Senior ML Engineer, Google DeepMind

I'm interested in bridging the gap between research and engineering. Over the past years I've been working on scaling AI training systems and frameworks. I'm also passionate about making sure that work stays aligned and safe as it scales.

"endless forms most beautiful and most wonderful have been, and are being, evolved" — Darwin, 1859

github ↗ twitter ↗ silicon tales ↗ writing →

Work

Google DeepMind

Senior ML Engineer · Gemini pretraining libraries and infra

Build and scale infrastructure for Gemini pretraining.
Build mid-training infrastructure for large-scale RL, focused on run stability at scale.
My latest paper: arxiv.org/abs/2605.23066 ↗

AWS

Scaling large multimodal models on Inferentia and Trainium

Cornell University

MEng, Computer Science

Writing

Essay · Nov 11, 2025

A History of Humanity's Quest for the Best Answer

A journey through the history of optimization, from Gauss's lost planet to the future of AI.

all posts →