I'm a senior AI engineer on the Enterprise AI team at Toyota North America. My foundation is
software engineering, and my work spans neural network training, AI agent development, cloud computing, systems
architecture, full-stack development, and API design.
I learn by building from first principles. I like knowing how things work at their foundation, so I've coded and trained large language models end to end, including a 124-million-parameter GPT-2 I trained from scratch on 10 billion tokens. I've also built neural networks in pure Python and NumPy, a ReAct agent from scratch that plays Microsoft's TextWorld (a reinforcement learning environment), and much more.
I'm also a big fan of mechanistic interpretability (essentially, neural network surgery), and
I study it closely. The answers to how these models compute are already inside them; we just
can't see them clearly yet. Some of my favorite work is running experiments: digging into a
model's internals, and studying its behavior the way you'd study an organism, watching what it
does in different scenarios and why.
I played soccer professionally in both Brazil and the United States. In
Brazil I played for Sociedade Esportiva Brazlândia in the Campeonato Brasiliense, and in the
U.S. I played for the Dallas Sidekicks in the Major Arena Soccer League.
I'm also a third-degree
black belt in Brazilian Jiu-Jitsu, promoted by my father, Master John Machado, one of the five world-renowned Machado brothers. Jiu-Jitsu is still a big part of my life today, and together with soccer, it
gave me the precision, focus, resilience, and creativity I bring to every team I work on.