About
I'm Kelvin Azevedo Santos — a founder and mechanism designer working on incentive mechanisms for coordination and cooperation. I've built and run real markets at scale, and I've been interested in AI alignment for over seven years. I'm based in São Paulo, Brazil.1
I came up through Brazil's STEM system: admitted to ITA — the country's most selective STEM university — ranked 33rd nationally as a high-school sophomore, and to the University of São Paulo for both Medicine and its interdisciplinary Molecular Sciences research program, which I left to join a high-frequency-trading fund.1 Along the way I picked up a few competition honors: silver at the Brazilian Informatics Olympiad, a reserve spot on Brazil's International Chemistry Olympiad team, first nationwide in the Fuvestão competition twice, and a CVM Professional Investor certification.1
I found LessWrong through Nick Bostrom's references to Yudkowsky's AI-box experiments, and somewhere along the way read Sutton & Barto's reinforcement-learning textbook cover to cover.2
I was born on October 21, 1995.3
I work mostly in mechanism design and markets and in applied ML — reinforcement learning, agentic AI, LLMs; I code in C, Python, and JAX. Portuguese is my native language; I'm fluent in English.1
Around
Outside of work, I'm a father of two — Elizabeth and Louise — and I travel a fair amount.
Me as a kid in Brazil.
With Robin Hanson (center) and Bryan Caplan (right).
My daughters, Elizabeth and Louise (faces blurred).
Bangkok.
Soup season.
Winter walk.