# Henrique Godoy

> Henrique Godoy is a Brazilian AI engineer and researcher based in São Paulo. He builds AI at CloudWalk, a $1.2B ARR fintech, across JIM (US) and InfinitePay (Brazil), serving millions of daily interactions across 6M+ merchants. He previously led the first production deployment of a large language model inside one of the largest investment banks in Latin America, and works on evaluation and reliability for AI systems operating in high-stakes environments. He founded Doki, a personal-finance product for doctors in Brazil that managed over R$10 million. He was admitted to the University of São Paulo (USP) mathematics program at 15 and ranked top 200 in the Brazilian Mathematical Olympiad for University Students (OBMU), the Brazilian counterpart of the Putnam Competition.

## Now

Henrique Godoy is building AI at CloudWalk — a $1.2B ARR fintech — across JIM (US) and InfinitePay (Brazil), serving millions of daily interactions across 6M+ merchants, with a focus on evaluation and reliability for language-model systems in high-stakes environments.

## About

- Name: Henrique Godoy
- Work: AI engineering and research; evaluation and reliability of LLM systems
- Location: São Paulo, Brazil
- Education: Computer Science at Inteli; Mathematics at USP (admitted at 15)
- Selected work: led the first production LLM deployment at one of Latin America's largest investment banks; founder of Doki (personal finance for doctors, R$10M+); 500K scholarship at Inteli; collaborations with Dell, Pirelli and Vivo.
- Mathematics: top 200, OBMU (Brazilian Putnam equivalent)
- Writing: superintelligence, leverage, systems thinking

## Interests

Reliability and evaluation of AI systems, applied mathematics, automating real-world processes, and the long-run trajectory of machine intelligence.

## Research

- Extract-0: A Specialized Language Model for Document Information Extraction (2025). A 7B-parameter model that outperforms GPT-4.1 and o3 on document extraction via synthetic data, LoRA fine-tuning, and GRPO reinforcement learning. Paper: https://arxiv.org/abs/2509.22906 · Code: https://github.com/herniqeu/extract0 · Data: https://huggingface.co/datasets/HenriqueGodoy/extract-0
- Alvorada-Bench: Can Language Models Solve Brazilian University Entrance Exams? (2025). A 4,515-question benchmark evaluating twenty language models on five Brazilian university entrance exams. Paper: https://arxiv.org/abs/2508.15835 · Data: https://huggingface.co/datasets/HenriqueGodoy/Alvorada-bench
- Full list with abstracts and BibTeX: https://henriquegodoy.com/research

## Links

- Website: https://henriquegodoy.com
- Profile (structured Q&A): https://henriquegodoy.com/ai
- Research: https://henriquegodoy.com/research
- Google Scholar: https://scholar.google.com/citations?user=LvmiOT0AAAAJ
- Hugging Face: https://huggingface.co/HenriqueGodoy
- GitHub: https://github.com/herniqeu
- LinkedIn: https://www.linkedin.com/in/henrique-godoy-879138252/

## Writing

- Why models write slop: the environments are too small — https://henriquegodoy.com/blog/why-models-write-slop
- Stream-of-Consciousness on Superintelligence — https://henriquegodoy.com/blog/stream-of-consciousness
- Lessons for My Past Self — https://henriquegodoy.com/blog/lessons-for-my-past-self

## Full profile

- https://henriquegodoy.com/llms-full.txt