Curriculum Vitae

A PDF version is available here.

Experience

Research Scientist, Google DeepMind: July 2026–present · New York, USA; Developing AI weather forecasting models for WeatherNext.
Research Scientist, Salient Predictions: December 2022–July 2026 · Remote (Croatia), then New York, USA; Developed probabilistic weather forecasting models across medium-range, subseasonal, and seasonal timescales.; Developed methods for combining forecasts from multiple models, including neural-network approaches.; Co-developed GEM-2 and GEM-3, efficient global weather models for probabilistic forecasting.
Research Resident, Cervest: June 2021–December 2021 · London, UK; Built probabilistic models of the urban heat island effect using geospatial data.
Research Analyst, Center for Computational Astrophysics, Flatiron Institute: February 2020–June 2020 · New York, USA; Developed a probabilistic model to reconstruct time-varying planetary surface maps from light curves using nonnegative matrix factorization and variational inference.

PhD in Astrophysics, University of St Andrews: 2017–2023 · St Andrews, Scotland; Developed Bayesian methods for gravitational microlensing and planetary surface mapping. Created caustics, an open-source JAX package for differentiable microlensing light curves.
MSc in Physics with Astrophysics, University of Rijeka: 2015–2017 · Rijeka, Croatia; Thesis research on circumbinary exoplanet dynamics at Lund Observatory, Sweden.
BSc in Physics, University of Split: 2012–2015 · Split, Croatia; Studied whether tidal engulfment explains the scarcity of close-in giant planets around evolved stars.

See the research page for a full list, or ADS.

Multi-model blending of probabilistic weather forecasts: US Patent Application 2026/0003100 A1 · filed June 2025; S. J. Levang, F. Bartolić. Machine-learning blender that combines the outputs of multiple probabilistic weather models with weights that adapt to location, lead time, and season.
Multi-model blending via a neural network for probabilistic weather forecasts: US Patent Application 2026/0003101 A1 · filed June 2025; S. J. Levang, F. Bartolić. Neural-network approach to probabilistic multi-model blending, learning context-dependent weights from location, lead time, and season.
Long-term weather forecasting framework using statistical and machine learning modeling: US Patent Application 2026/0104533 A1 · filed October 2025; S. Ridge, V. Cikojević, F. Bartolić. Hybrid framework combining statistical dimensionality reduction with a diffusion model for long-term forecasts.