Born in Rio, based in Paris. My first research, as an undergraduate at UFRJ's Institute of Physics, was on the logistic map: how a one-line equation slips between order and chaos. That led me to fluid mechanics, from rising bubbles at ESPCI to turbulence models at ONERA, and through data-driven turbulence modelling to machine learning.
Today I lead the AI and ML behind Calldesk's voice agents: LLM agents, retrieval, the models we train in-house, and the evaluation that tells us whether any of it really works. Outside work I'm drawn to civic tech, and I photograph, mostly in the street and mostly in black and white.
In the first round of Brazil's 2026 presidential election, 6 in 10 Brazilians living abroad didn't vote, largely because of distance: 164 polling places for 102 countries. Between the two rounds, I built a non-partisan noticeboard where people offered a seat in their car or a bed for the night to voters far from their polling place.
Next.js, Postgres with row-level security and 126 database tests, privacy by design, and electoral law turned into product requirements. Covered by TV Cultura, Folha de S.Paulo, DW and others.
votolafora.com.br → code →Technical owner of every AI and ML feature as the company's only ML specialist, from research spike to production: AI engineering (LLM agents, prompts, retrieval) and classic ML (in-house models trained on production data) for voice agents that take real phone calls.
Health-insurance AI. Led a team of five responsible for document classification, OCR, named entity recognition and medical coding: set the R&D direction of each project, ran sprint planning and kept the work aligned with business and technical goals.
In the AI research division, designed NLP, computer vision and multimodal models to extract information and structure from the documents of health insurers.
Data-driven turbulence modelling for the aeroelasticity of laminar wings: laminar separation flutter at transitional Reynolds numbers, turbulence models corrected by data assimilation (variational field inversion) under the Navier–Stokes equations, then generalised to unseen configurations with deep learning. Advised by Vincent Mons and Olivier Marquet.
How a bubble rises when gravity changes quickly in time: numerical and theoretical modelling of the interface, and how the cell's geometry and the bubble's size lead to toroidal bubbles and chaotic behaviour. Simulations in Basilisk.
Computational fluid dynamics with OpenFOAM for the oil and gas industry, alongside the company's engineers.
Statistical models of erosive wear: built a wear database and a simulator comparing empirical models, and wrote the technical reports.
My first research: the critical behaviour of the logistic map's bifurcation diagram where chaos gives way to a periodic window, and the map extended to the complex plane to study its stability.