AI/ML & Energy | Geoscientist Senior Leader | E&P, Electricity, Renewables & Storage | Portfolio Strategy & Capital Allocation | AI Governance
I build AI systems that turn forecasts into decisions an energy company can defend: which projects to fund, how to dispatch an asset, when to alert an operator. My background is twenty years in the energy industry, from subsurface geoscience to portfolio strategy and enterprise AI, and an M.S. in Applied Data Science (AI concentration) from the University of Michigan.
Based in the United States. U.S. permanent resident; German and Brazilian citizen.
Every project below runs on public data, ships its tests, and states its limitations.
| Project | What it does | Result |
|---|---|---|
| Capstone: Beyond Prediction | Risk-aware capital allocation across oil, gas, power, renewables and storage: survival models for interconnection queues, 2,000 joint price scenarios, a mixed-integer optimizer on expected NPV minus worst-5% loss, and a governed language-model challenger. University of Michigan capstone. | On a $3.82B budget and 40 candidates, the risk-aware book beats NPV ranking by $219M on held-out scenarios (modeled, 95% interval +$13M to +$401M). |
| forecast-to-dispatch | Probabilistic ERCOT price forecasting and dispatch of a 100 MW / 200 MWh battery, settled at realized prices. | 86.1% of perfect-foresight revenue over 57 out-of-sample days. |
| netload-forecasting-flexibility | Hourly net-load forecasts, 1 to 72 hours ahead, for six U.S. balancing authorities, with calibrated uncertainty bands for reserve sizing. Team project. | Day-ahead error 30.1% lower than the EIA benchmark on held-out 2024 data; 90% band covers 88.7% of hours. |
| ai-enablement-proof | A governed agent that monitors U.S. grid stress and drafts operator advisories, with every action passing a coded governance gate. | Alert 11.9 hours before ERCOT's EEA3 in Winter Storm Uri and 9.5 hours before PJM's alert in Elliott; false-alarm rate published. |
| cmapss-rul-governance | Remaining-useful-life prediction on NASA C-MAPSS with drift monitoring, rollback and human override. | Governance mapped to NIST AI RMF 1.0 and the EU AI Act, enforced in CI. |
- Decision first. A model is judged by the decision it improves, in dollars and risk, not by its accuracy alone.
- Held-out evidence. Results are scored on data the model never saw, against the strongest simple baseline I can build.
- Governance in code. Limits, approvals and audit trails are tests that block a release, not slides.
- Limits stated. Each project says what it does not show.
- Energy. Twenty years at Petrobras across geoscience, petrophysics, reservoir studies, portfolio management, energy transition and enterprise AI.
- AI and data science. Forecasting, survival analysis, optimization under uncertainty, generative AI with human oversight.
- Strategy. Capital allocation, risk management and leading multidisciplinary teams in capital-intensive businesses.
- M.S. Applied Data Science, AI concentration, University of Michigan (2026)
- M.Sc. Reservoir Geoscience and Engineering, IFP School, France
- Executive MBA in Strategy, COPPEAD / UFRJ
- Executive Program in General Management (EPGM), MIT Sloan School of Management
- General Management Program (GMP), Harvard Business School
- Renewable Energy and Emerging Technologies, Stanford University
- Data Engineering Certificate, MIT
- B.S. Geology, UFPR
I write on AI, energy and executive decision-making on Medium.