The math behind keeping Brazil's lights on.
Brazil's power grid runs mostly on water — roughly two thirds hydroelectric — so operating it means deciding how much to release from the reservoirs now and how much to save, without knowing when it will rain. My doctorate (FGV/EMAp) works on the stochastic optimization behind that decision: SDDP, cut selection, and the NEWAVE–DECOMP models the national operator actually runs in production.
Stochastic Optimization
SDDP, decomposition methods, optimal control. Applications in hydrothermal dispatch and energy planning, and everywhere else i can find.
PhD · FGV/EMApGenerative Models
Normalizing flows, continuous NFs, score & diffusion models, flow matching.
MSc · COPPE/UFRJAI-Agent Engineering
Multi-agent SDLC pipelines, guardrails, RAG, type-aware AST mapping over large codebases, context engineering! I love making agents think!
Pluma FinanceScientific Computing
Numerical methods for ODEs/PDEs, CFD, diffusion-based super-resolution of reduced-order models.
NACAD/UFRJBuilding systems where mathematics matters.
Data engineer and general consultant at LabMA/UFRJ, helping generate and optimize the brazilian biometric tables. Earlier: Early engineer at Pluma Finance, quant trading systems and scientific ML.
Things I build and write, outside the day job.
Open-source tools — db_orchestrator, citation-xray, sparring, llm-chunking — and writeups on the same stuff. Mostly Python, mostly born from an actual itch.
Tábuas: optimizing a 1TB Oracle pipeline
The techniques that took an analytical pipeline from naïve to fast. Optimized bitmasks, plan analysis, and everything else.
coming soonMinions: an affordable multi-agent SDLC
Lessons on making a multi-agent software pipeline both cheap and functional. Real failure modes, and everything i learned about agents!
coming soon