AI-assisted forecasting
Investigating how machine learning can support weather prediction while keeping uncertainty, evaluation, and operational usefulness in view.
- Weather prediction
- Uncertainty-aware evaluation
- Human-centered forecasting
Research
Machine learning for weather—and the systems that make it trustworthy.
My work focuses on AI-assisted forecasting, rigorous model evaluation, and the practical infrastructure needed to turn an experiment into reliable evidence.
Research focus
Machine learning systems
Building reliable training and evaluation workflows for data-intensive scientific problems.
- Reproducible experiments
- Scalable data workflows
- Model diagnostics
Research tooling
Turning repetitive research tasks into practical tools that make complex work easier to run, inspect, and trust.
- Data pipelines
- Experiment tracking
- Workflow automation
How I approach the work
Build, measure, understand, repeat.
- 01
Start with the decision
Define what a useful result means before choosing a model or headline metric.
- 02
Make the workflow reproducible
Treat data preparation, experiment configuration, and progress evidence as part of the research.
- 03
Interrogate the result
Look beyond rank order to understand error patterns, tradeoffs, and where the system remains uncertain.
Current work
Active research
AI-assisted weather forecasting
Exploring machine-learning approaches to weather prediction, with an emphasis on rigorous evaluation and practical forecasting value.
Machine learningWeatherEvaluationResearch systems
In development
Reproducible research workflows
Practical tooling for moving from raw scientific data to repeatable experiments and inspectable results.
Data pipelinesAutomationExperiment tracking
Ongoing
Model evaluation and diagnostics
Evaluation practices that make model comparisons clearer, more honest, and more useful than a single headline metric.
DiagnosticsError analysisModel comparison
Publications & presentations
Work in progress.
Peer-reviewed publications, preprints, talks, and posters will be added here as they are ready to share. No placeholder citations—only real work.