AI & ML
AI that is checked against the physics of the equipment it describes.
I apply AI and machine learning to energy equipment, materials and industrial processes, where off-the-shelf methods struggle: little data, safety limits, or the need for predictions that make physical sense.
AI and ML are not a separate silo here: they run through my modeling, experimental, training and advisory work. There are three ways to engage on AI/ML itself — consulting, advisory and training.
What I do
Consulting
- Faster stand-ins for heavy simulations — I train fast models on a limited set of detailed simulations (methods such as reduced-order models and neural operators), then check them against the underlying physics.
- Machine learning on real operating data — spotting anomalies, planning maintenance and tuning processes, using data from energy equipment and plants.
- Screening materials and devices with data — ranking candidate materials or device designs early, to focus costly simulation and experiments.
Advisory
- Is AI worth it here? — helping technical and non-technical leaders judge where AI genuinely adds value and where it is hype, before committing budget.
Training
- AI/ML for technical teams — introductory courses and hands-on sessions, including physics-based methods built around a single worked example. See Education & Training.
Reach & application
Engagements run from single technical studies to ongoing advisory, and from hands-on model-building to C-level framing of an AI roadmap. Core domains include renewable energy, energy storage and conversion devices, and industrial process systems.
Why work with me
Every model I hand over is checked against the physics it’s meant to respect — not just against a validation split.
Illustrative use cases
- Faster simulations with a surrogate model — a reduced-order or neural-operator surrogate trained on a limited set of full simulations, giving fast approximate predictions for early design iteration.
- Predictive maintenance for energy systems — models on operating data from energy devices, grounded in known physical failure modes rather than pattern-matching alone.
- Screening candidate materials — data-driven ranking of candidate materials or device configurations (e.g. for alpha/betavoltaic or superconducting applications) ahead of detailed simulation or experiment.