SYSTEM_ONLINE
I build and validate weather & climate AI at petabyte scale.
A decade of geophysical modelling (Cambridge PhD, Stanford MSc), now benchmarking the AI models the weather and climate industry runs on.
- marEx PyPI downloads230k+as of 2026-07
UK Met Office, ECMWF & NOAA
- ResNimbus hackathon result2nd place, Climate Informatics 2026
- AI weather emulators benchmarked6
- Live products in production4
- Publications10 peer-reviewed, 7 first-author
The mesh is the icosahedral grid graph-based weather models compute on.
Everything I model is one coupled system.
A shock in the atmosphere becomes a change in the ocean, a swing in power generation, and a loss on a balance sheet. The models I build trace how one variable drives the next.
The connected system
Follow a thread from a climate variable to the energy and risk it drives. Hover a node, or focus one from the list, to trace what it influences.
Complex, nonlinear, interconnected: I build models of exactly this.
CASE_STUDY Cascade
Weather-shock propagation across the European energy graph, from a cold snap through
grid stress to insured loss.
Cascade stress-ranking AUC0.738 vs 0.652 persistence View case study FEATURED_WORK
Four threads through the system.
-
tailspec Tail-risk evaluation of the AI weather emulators the industry runs on. View case study -
Cascade Weather-shock propagation across the European energy graph. View case study -
ResNimbus Diagnosing cloud cover from AI weather models, calibrated across the domain shift. View case study -
marEx Open-source marine-extreme detection and tracking at petabyte scale. View case study
LIVE_PRODUCTS
Running in production.
4 in production, plus the tailspec evaluation dashboard.
CURRENTLY
Where things stand.
- Role
- EERIE climate-model evaluation lead, ETH Zürich
- Collaboration
- Machine-learning research with Oxford
- Based
- Zürich, Switzerland
- Right to work
- CH B permit · UK no sponsorship required