Experience
Experience and education
Research positions in machine learning, optimization and control, preceded by industry work on decision models under uncertainty and on engineering simulation. The full CV is available as a PDF.
Research and industry
-
Graduate Visiting Researcher
Robot Learning Lab, UC Berkeley · with Pieter Abbeel
Policy learning and distillation for humanoid locomotion and dexterous manipulation.
-
Graduate Researcher
Los Alamos National Laboratory, Los Alamos, NM
Decision models inside learning pipelines: Solution sensitivities of nonconvex programs (DiffOpt.jl), optimization proxies trained with selected sensitivities, and strategic bidding through AC market clearing, with energy and robotic systems as test cases.
-
Graduate Researcher
NSF AI Institute for Advances in Optimization (AI4OPT), Georgia Tech, Atlanta
Lead of the LearningToOptimize open-source organization and co-designer, with Michael Klamkin, of the student-led PhD course ISyE 8803.
Policy learning for constrained sequential decisions (two-stage generalized decision rules), learned value functions for optimal power flow, and calibration of simplified decision models.
-
Researcher
Invenia Labs, Cambridge, UK
Risk-aware decision models under uncertainty for bidding in seven US electricity markets (MISO, PJM, ERCOT, CAISO, SPP, ISO-NE and NYISO). Related research: estimating uncertainty for robust decisions and strategic bidding.
- Directed projects on daily bidding under price uncertainty, in which machine-learning estimates of that uncertainty define robust and distributionally robust bidding models.
- Contributed to a detailed market simulator for counterfactual analysis, and led the data analysis of simulation accuracy.
- Developed decomposition methods for large-scale integer and continuous models, and estimated unknown grid parameters with inverse optimization and heuristics.
- Explored exact and surrogate methods for bilevel problems such as strategic bidding and hyperparameter tuning, using automatic differentiation and automatic dualization of optimization models.
-
Research Staff Member
LAMPS, Pontifical Catholic University of Rio de Janeiro · consulting projects
- Mathematical-programming models for optimal transmission-usage contracting for Energisa Group, one of Brazil’s largest electricity companies.
- Flexible software for multistage hydrothermal economic dispatch in a consulting project for FGV Energia, assessing network and capacity changes (HydroPowerModels.jl).
-
Intern
Banco BBM (BOCOM BBM), Rio de Janeiro
Fraud-risk assessment of trade notes receivable in the Data Science team; macroeconomic indicator analysis in the Economic Research team.
-
Research Intern
Air Liquide Medical Systems, Paris
Built the company’s first integrated simulator of a breathing-support machine, covering the CPAP and BiPAP programs of the Vendôme home ventilator, so researchers could test new designs.
-
Intern
Northumberland, Tyne and Wear NHS Foundation Trust, Newcastle, UK
Digital consumables database for stock control, and protocols for logging supply usage.
Education
-
PhD in Machine Learning
H. Milton Stewart School of Industrial and Systems Engineering, Georgia Tech · AI4OPT Fellowship
Advisor: Pascal Van Hentenryck. Thesis topic: policy learning for constrained sequential decision-making.
-
MSc in Operations Research (Electrical Engineering)
Pontifical Catholic University of Rio de Janeiro (PUC-Rio)
Advisor: Alexandre Street; co-advisor: Davi Valladão; committee: Andy Philpott, Arild Helseth, Bruno Fanzeres. Thesis: A Framework for Assessing the Impacts of Network Formulations in the Operation of Hydrothermal Power Systems. Research on decision-making under uncertainty in quantitative finance, including portfolio selection and asset–liability management, and in energy systems.
-
Double degree in General Engineering · BRAFITEC merit scholarship
École Centrale de Marseille, France
Minor in computer science and information systems. Project: EEG-based epilepsy classification at the Fresnel Institute.
-
BEng in Mechatronics and Control Engineering
Pontifical Catholic University of Rio de Janeiro (PUC-Rio)
Project: Aircraft modeling and microcontroller programming with the AeroDesign team.
Technical computing and languages
Open-source systems
Author of Julia packages behind the research on this site and contributor to DiffOpt.jl in the JuMP ecosystem. The software work spans optimization modeling, solution sensitivities, multistage planning, and GPU-oriented workflows.
Programming languages
Julia and Python are central to current work. Earlier and continuing technical work includes C, C++, C#, Java, and SQL, spanning numerical software, object-oriented programming, data systems, and embedded computing.
Scientific and engineering tools
JuMP, DiffOpt, automatic differentiation, GPU solvers, and PyTorch for current optimization and learning systems; MATLAB for EEG classification; SolidWorks and microcontroller programming for mechatronic design.
Human languages
Portuguese (native) · English (full professional) · French (full professional, DALF C1) · German (B1, DSD I)