Infinigrid
Production ML for the Norwegian power grid. Forecasting models that watch the electrical system and predict its risks before they bite. Time-series at grid scale, deployed.
machine learning · hardware · systems
Machine learning engineer with hardware roots. 14+ years experience across production systems and hobby projects. Built on clean code and robust programs.
I build
Three to look at first.
Production ML for the Norwegian power grid. Forecasting models that watch the electrical system and predict its risks before they bite. Time-series at grid scale, deployed.
Industrial optimisation for one of the world's largest aluminium producers. A mix of machine learning, classical optimisation, and the applied math that holds the two together.
Adapting a pre-trained vision transformer to a new task with low-rank adapters in the attention layers. Half make-it-work, half an excuse to actually understand why it works.
The tools I work in, grouped by depth.
A few you can try right now, then the rest grouped by theme.
Have an AI model predict your gender, age, and even mood from the colours you prefer.
try it hereMultiple AI content channels across several social platforms: automatic video generation and publishing.
see an exampleFlappy Bird where your voice is the controller: the louder you are, the higher the bird climbs.
play it hereTwo-player three-in-a-row where bigger blobs swallow smaller ones, and what is underneath stays hidden.
play it hereA custom autonomous-capable boat: designed hull, lidar, dual gimballed cameras, and onboard buoy detection.
see the buildAn RC car streaming dual cameras over 4G with a custom hole-punching protocol, drivable from anywhere.
watch it driveProjects I have built and milestones along the way, from 2002 to now.
Size = significance. Tap a project for the deep-dive.
The timeline only shows some of the projects I've finished. Like most engineers, starting something new is often more exciting than finishing the old, and a lot of what I know lives in the projects that never quite got there.
The FPGA design that synthesised but I never tested on hardware. The audio plugin I abandoned the moment I'd learned the trick. The model I trained until the loss curve told me what I needed, then walked away from. They're not failures. They're the experiments that taught the techniques that ended up in the projects that did ship.
Who, what, and why the two halves fit together.
I started writing code at nine. My father is a software developer, and he handed it over early. The first thing I remember being mad about was a coding class that turned out to be drag-and-drop blocks instead of a real keyboard.
Today the work has three loops. Machine learning for the Norwegian public sector, industry, and now Infinigrid, a startup building forecasting models for the electrical grid. Full-stack sharpened in Go and Nuxt: backends, dashboards, pipelines. Hardware: soldered audio front-ends, FPGA filters, a kit-built 3D printer that eventually drew its own portrait.
The two halves feed each other. The CNN that classifies an analog signal is only as good as the op-amp in front of the ADC; the FPGA that filters a transducer only matters if there's a pipeline downstream. I like working where that boundary lives.
Tools by where they sit in the chain.
From physical signal to clean data.
Where the learning happens.
Backends, infra, the unglamorous spine.
The surface a user actually touches.
currently going deeper on grid-scale forecastingcausal inferencecontrol systems
Open inbox.
Happy to talk research, hardware, or grid-scale ML, especially where the signal-chain crosses the model boundary.