Maike Holthuijzen

I’m an Applied Machine Learning Scientist with 5+ years of experience building and deploying predictive and anomaly detection models for large-scale real-world systems, especially in energy, Earth systems, and climate. My work centers on feature engineering, probabilistic modeling, Bayesian inferences, and translating model outputs into actionable decision support.

At Sandia National Laboratories (2023–Present), I developed and operationalized ML/DL models for climate- and energy-related natural hazard risk assessments, including Random Forest, XGBoost, and ESN for subseasonal multi-region forecasting. I design Gaussian process and deep learning surrogate models for Bayesian calibration and optimal experimental design, develop novel conformal prediction techniques for spatially and temporally dependent data to add uncertainty quantification, and build graph neural network approaches for heat-hazard impacts on electrical grids.

Previously at Virginia Tech (2022–2023), I developed a statistical method for forecasting lake temperature profiles using Gaussian process surrogates, and I’m a first author in Annals of Applied Statistics. I also create tools like the R package hetGP4cast and, in practice, I integrate and deploy end-to-end pipelines—combining anomaly detection with local LLMs (via Ollama), using Chainlit for user interaction and interpretability, and deploying Dockerized microservices with Anomalib.