About.
I'm an AI Engineer working in production AI, building and evaluating LLMs, agentic architectures, evaluation frameworks, and applied ML.
On the research side, I'm a Research Fellow at Black in AI Safety & Ethics (BASE Research), collaborating with Serena Oduro (Data & Society Institute) on assessing AI safety evaluation practices and whether they meaningfully address safety concerns impacting Black communities. I'm also an AI Safety Research Fellow at SPAR, working on efficient LLM safety benchmarking with Item Response Theory.
I hold an MSc in Computer Science from UFMG, with research focused on ML fairness and bias mitigation, plus a B.Sc in Computational Mathematics from the same university.
Research interests.
On the research side, I work at the intersection of AI fairness, AI safety, and the social sciences, questions like how bias enters ML pipelines, how to mitigate it in practice, and how AI systems land in real-world social contexts. On the engineering side, I focus on large-scale model evaluation, ML reliability and robustness, and applied AI prototyping.
- Fairness & bias mitigation
- AI safety & AI for good
- LLM evaluation practices
- AI ↔ social sciences
If you're working on these or related topics and want to collaborate, please reach out.
Background.
Experience 5+ yrs
-
Staff Data Scientist & AI EngineerNow Apr 2026 →
Flash Technology · São Paulo / Remote
Production AI for financial applications, working with LLMs, agentic architectures, evaluation frameworks, applied ML.
-
Research Fellow · AI Safety & EthicsNow May 2026 →
BASE Research · Part-time, Remote
Working with Serena Oduro (Data & Society Institute) on assessing the current state of AI safety evaluation practices and their ability to address safety concerns impacting Black communities.
-
AI Safety Research FellowNow Feb 2026 →
SPAR Research · Part-time, Remote
Efficient LLM safety benchmarking via Item Response Theory (IRT) & adaptive item selection. The method achieve up to 95% lower eval cost while preserving ranking accuracy.
-
AI Engineer · AI Models Team Oct 2021 — Jan 2026
IBM Research · Brasil / Worldwide
Led innovation initiatives on the proprietary Granite foundation models, such as, model quality, evaluation pipelines, and enterprise adoption. Close collaboration with the MIT-IBM Watson AI Lab on LLM evaluation, validation, prompt engineering, and exploratory science research.
Education & Research 3
-
MSc · Computer Science 2022 — 2024
UFMG · Federal University of Minas Gerais
Thesis on ML fairness in practice, a framework for designing bias-mitigation interventions. Advised by Ana Paula Couto, Marisa Vasconcelos, Mariano Beiró.
-
Visiting Scholar & Research Fellow 2022 — 2023
UC Berkeley · BEAAMO / BAIR
Auditing algorithmic & evidentiary statistical software used in criminal legal systems. Advised by Prof. Rediet Abebe.
-
B.Sc · Computational Mathematics 2013 — 2019
UFMG · Federal University of Minas Gerais
CS + applied mathematics. Undergrad researcher on Hidden Markov Models for speech recognition & genome sequence analysis.
Open-
source
- grafite IBM Research framework for analyzing LLM regressions across releases. Co-author.
- ibm-generative-ai Python SDK for IBM's foundation-model API. Initial author & maintainer (≤ v0.2.6).
Service
& community
- Black in AI · NeurIPS Workshop Co-Chair2020 – 2024
- Black in NLP · NAACL Core Organizer2021
- Girls Support Girls · Founder2018 – 2022
- LatinX in AI · ICML Volunteer2020
- Women in ML · ICML Volunteer2020
Off the clock.
I like to travel, play with the lenses on my phone as if I were actually a photographer (see some here), buy more books than I read, and watch superheroes and horror movies — with a subscription to every stream so I don't miss any TV shows. I don't actively consume YouTube, but I enjoy videos like 3Blue1Brown.
May the force be with you.Preprints.
Under review · not yet peer-reviewed-
Grafite: Generative regression analysis framework for issue tracking and evaluation
arXiv:2603.18173 [cs.CL] · Open-source project (IBM/grafite)
Publications.
Peer-reviewed-
Efficient Multiprompt Evaluation of LLMs
The Thirty-Eighth Annual Conference on Neural Information Processing Systems
-
LLM Routing with Benchmark Datasets
NeurIPS 2023 Workshop on Distribution Shifts: New Frontiers with Foundation Models · Oral presentation
Master's thesis.
UFMG · 2024-
Master's Thesis · Federal University of Minas Gerais (UFMG), Belo Horizonte, Brazil
Poster presentations.
Workshops-
ICML 2024 Workshop on Humans, Algorithmic Decision-Making, Society · Poster presentation (non-proceeding)
-
Biological Sequence Analysis using Profile Hidden Markov Models
NeurIPS 2019 Black in AI (BAI) Affinity Workshop · Poster presentation
-
O modelo certo para a tarefa certa: estratégias de roteamento para LLMs de código aberto
Bradesco Inovabra · IBM Research
-
AI, academia, and industry: bridging the gap between theory and practice
Women in Data Science (WiDS) Conference · Guayaquil @ ESPOL, Ecuador
-
AI, ML & Deep Learning: what's the difference, and how is algorithmic bias generated?
PrograMaria Encontros · powered by Avanade
-
Data-centric AI and feature stores: data-centric development for performance, quality & selection
Serasa Experian DataLab