Research
My scientific interests have been varied. I have published papers on turbulence theory, fusion optimisation, archaeological science and machine learning.
Current work: Power Factors (formerly Greenbyte), 2018 – now
At Power Factors I work on agentic AI, machine learning and the platforms that let data scientists build and run their own work at scale, and I have owned the battery energy storage roadmap for our Asset Performance Management (APM) product since 2019.
Agentic AI
- Led my team’s move to agentic AI, with a measured gain in productivity.
- Jointly led company efforts to embrace agentic AI, and set up a company-wide AI plugin marketplace.
- Built a shared RAG service and MCP server for long-term agent recall.
Hybrid modelling and machine learning
- Built a recommendation engine for wind predictive maintenance; mentored colleagues building digital twins for solar farms, and a Chalmers master’s student whose autoencoder work was published in Renewable Energy (2020).
Platforms and teams
- Designed and built the calculation platform behind many of our analytics, used by several teams.
- Led the building of a GraphQL API giving structured access to data from solar, wind and battery assets across our customers’ portfolios.
- Recruited and trained a new 8-person MLOps team so that data scientists could build and run their own work.
Battery energy storage
- Recognised early that grid-scale battery storage would matter, and helped build Power Factors’ storage offering from the ground up.
- Owner of the APM product roadmap for battery energy storage systems since 2019, built around close work with our largest storage customers; led the storage components of APM.
Talks
- Data Science with evolving data — GAIA meetup, Data Science for Renewable Energy, 2019: a data-version-aware workflow on the open-source Luigi framework. Video
Archaeological science, with Amber Hood
I have also worked with the archaeological scientist Amber Hood on dating ancient Egyptian pottery by luminescence, bringing the modelling and mathematics from my physics background. All three papers are open access.
Relative dating without the burial environment. Luminescence dating measures the radiation a pot has absorbed since it was fired, and normally needs to know how much radiation came from the ground it was buried in. For many museum pots that information is lost. In Part A we derived a formula for the relative ages of two vessels that does without it, and showed that the resulting error is much smaller than for an absolute age: around 3.6% rather than 45% in our example. In Part B we tested it on wavy-handled vessels from Turah and wine jars from Turah and Hierakonpolis, from about 3300–3000 BC, and found the sequence largely in keeping with the established pottery typology.
Reconstructing tombs in 3D. The radiation a pot receives in a tomb depends on its surroundings: bedrock, mud-brick, fill and empty space. In Using DosiVox to Reconstruct Radiation Transport through Complex Archaeological Environments (Methods and Protocols, 2019) we rebuilt tombs at Bêt Khallaf and the Tomb of Djer at Abydos as 3D models in DosiVox, a particle-physics simulation, to calculate that dose. In the Tomb of Djer the result differed significantly from the usual simplified geometric estimates. I carried out the computational modelling and its validation, and derived the geometric approximation in the appendix.
A ceramic cache in a chamber tomb, and the four most common ways a vessel lies in the fill; the red crosses mark where OSL samples are taken. Figure 5 of Hood and Highcock (2019), Methods and Protocols 2(4), 91. CC BY 4.0.
The DosiVox model for a vessel in the Tomb of Djer: the tomb is divided into voxels of mud-brick (white), fill (green) and air (blue), around the vessel walls (red) and base (cyan). Figure 4 of the same paper.
Earlier: fusion research, 2007 – 2018
The rest of this page is adapted from the 2014 version of the site.
Before moving to industry I worked on plasma turbulence in fusion reactors, at Oxford (PhD 2007–2012, then fellowships to 2016) and Chalmers (2016–2018). I carried out the first optimisation of tokamak fusion reactor performance using nonlinear turbulence simulations, built the international collaboration needed to run it on GPU-powered supercomputers, and demonstrated that rotational shear could suppress turbulent heat loss.
Why turbulence matters for fusion
Fusion is a way of producing energy by colliding two isotopes of hydrogen (deuterium and tritium) to make helium and a neutron. The challenge of getting fusion to work is the challenge of keeping the fuel hot enough for the reactions to happen. The fuel is too hot to put inside a container, so it is held in a very powerful magnetic field. Even when that magnetic “cage” has been made stable, turbulence can let the heat leak out and prevent the fuel from getting hot enough. Turbulence is a chaotic mess of fluctuations and whirling vortices which flings the heat out from the plasma.
Suppressing turbulence
My work was focused on eliminating turbulence. The figure below shows a calculation of the highest temperature that can be achieved without generating turbulence (the “zero-turbulence manifold”).
Subcritical turbulence
In our usual picture of turbulence, small fluctuations within the plasma are driven unstable by the enormous pressure gradients, and grow to become strong waves, eventually interacting with each other and breaking each other up to form turbulence. However, sometimes one can get turbulence even if the small fluctuations are stable. This is known as subcritical turbulence. I investigated subcritical turbulence within a plasma and tried to determine when it is present.
Video: formation of turbulence.
Tools to simulate turbulence
The turbulent vortices that convect heat out of the plasma are elongated, following the magnetic field lines which wrap around the toroidal plasma. We use a coordinate transform (pictured in the video) to compress these structures into a numerical box so that simulations of them become possible. Over the past two decades the tools used to study these vortices have improved beyond all recognition: we can now simulate, within a few hours, the effects of fully saturated nonlinear turbulence. The challenge is to do that in a matter of minutes, so that the effects of turbulence can be investigated in many different scenarios.
See the code projects for the simulation tools, the publications for the papers, and the collaborators page.
Related links
- Collaborators — Co-authors and collaborators, as listed on the 2014 site.