Machine Learning · Climate Science · Forecasting
I build production ML systems at the intersection of climate science and operational forecasting.
Machine Learning engineer and PhD researcher with five-plus years building production-grade ML across academia, public sector and industry.
I'm currently completing my PhD in Climate Science & Machine Learning at Universidad Complutense de Madrid as an FPI Fellow (Spanish State Research Agency; sub-5% national acceptance rate, defending Q1 2028), supervised by Prof. Javier Martín-Chivelet and Dr. Mario Morellón.
My work focuses on turning complex, high-volume environmental data into operational decisions — from karst-water forecasting in Switzerland to atmospheric attribution across southwestern Europe. Recently designed Tempestas, an original probabilistic deep learning framework, during a six-month research stay at the University of Michigan.
Lead author on four peer-reviewed manuscripts at Q1 climate journals (Weather and Climate Dynamics, Global and Planetary Change); regular presenter at EGU, EMS and Karst Records. Currently serving as Expert Reviewer for the IPCC AR7 Working Group I First Order Draft (Chapter 6), through the Early Career Researcher Group Review programme.
Outside the lab, a speleologist for over a decade — patience, resilience, and a deep appreciation for how complex natural systems actually behave.
A selection of projects spanning operational ML, probabilistic deep learning and applied climate science.
An original probabilistic deep learning framework: Mixture-of-Experts architecture in PyTorch with a custom Gaussian NLL loss, providing explicit treatment of aleatoric and epistemic uncertainty for high-dimensional spatiotemporal modelling.
A three-stage XGBoost cascade producing hourly forecasts at nine lead times (one hour to seven-day horizon) on 30+ years of operational data. Real-time deployment integrates ECMWF and ICON ensemble forcing with regime-dependent bias correction.
Predicts solar and wind power output, forecasts electricity demand, and recommends the optimal 24-hour generation mix by combining ML forecasting with linear optimisation.
View repository →Machine-learning attribution of synoptic-scale climate variability (NAO vs. WeMO) on rainfall isotopic composition in southeastern Iberia, combining Random Forest, SHAP and HYSPLIT trajectory clustering.
Zenodo record →A reproducible Random Forest framework quantifying the coupling between large-scale atmospheric circulation (ERA5, IsoGSM, NAO, WeMO) and rainfall δ¹⁸O across southwestern Europe.
Zenodo record →Tools I reach for on a daily basis.
Peer-reviewed publications and conference contributions.
EGUsphere preprint · Under review at Weather and Climate Dynamics (Q1)
Random Forest · SHAP · HYSPLIT trajectory clustering
Under review at Global and Planetary Change (Q1, Elsevier)
Random Forest · Permutation feature importance · ERA5 · IsoGSM
KR10, South Africa
European Meteorological Society (EMS) Annual Meeting
European Geosciences Union (EGU) General Assembly
S4 Speleothem Summer School, São Paulo
Master's research co-supervision and support lectureships at Universidad Complutense de Madrid.
Member of the organizing committee of the S4 Summer School (Morocco, 2025), a multi-day international event connecting senior researchers with early-career scientists working on cave and climate records.
speleothemschool.com →Active speleologist for over a decade. The curiosity that drives my ML work on cave-derived climate records started underground, where you learn to read complex natural systems patiently and with respect for what's unknown.
Open to roles in climate-tech, energy transition and forecasting-heavy industries. Always happy to chat about ML, uncertainty quantification or applying analytics to messy real-world data.
