About
I'm the CTO and co-founder of Avra, where we train graph-native foundation models on large-scale economic networks. Most of my current work is on learning representations from relational structure — how entities connect through ownership, legal proceedings, and geography — and the inductive advantages over feature-engineered baselines.
More broadly, I'm interested in self-supervised learning on graphs, representation learning, and the problems that come with training on heterogeneous, temporal data at scale.
Earlier, I was part of OpenAI's SOCML research program (2016–2018), where I worked on recurrent networks, time series, and generative models for text. I'm a founding member of ABRIA, the Brazilian Association of Artificial Intelligence. Before Avra, I led AI at Jusbrasil and co-founded Leadster (CTO, seven years). University of London dropout.