I bridge heavy enterprise data and elite GPU performance. Energy grids, industrial digital twins and real-time risk — rendered at 60 FPS in the browser.
A live MapLibre + deck.gl scene rendering 1,000,000 synthetic grid-asset points over Norway. All geometry uploaded once as binary GPU buffers — pan, zoom and animate with zero CPU re-tessellation.
Drag to pan · scroll to zoom · color = simulated grid load (green→red). Watch the FPS counter hold steady while the point budget changes.
Hundreds of thousands of particles advected through an analytic flow field on a WGSL compute shader — modelling transfer from northern hydro generation to southern load centres. Trails accumulate in a ping-pong texture. The CPU writes one uniform per frame; everything else runs on the GPU.
This demo needs a WebGPU-capable browser (Chrome/Edge 113+, or Safari Technology Preview).
The compute-shader source ships in the bundle either way — open
src/webgpu.ts to read the WGSL.
Real @compute WGSL kernel · 64-thread workgroups · GPU-resident particle state.
A live exposure surface for fintech / insurtech. A 73k-triangle grid is displaced entirely in a custom GLSL vertex shader by an animated risk field — asset clusters that reprice every frame. The fragment shader ramps colour by exposure and shades with a screen-space normal. Zero CPU geometry work; drag to orbit.
Surface height & colour = simulated risk · repriced per frame on the GPU.
A real-time isochrone over real Trondheim terrain (Estenstadmarka / Jonsvatnet). A WebGPU compute shader solves the anisotropic eikonal equation on a 455k-cell DEM — hiking time from any point, with cost from Tobler's hiking function (uphill slow, gentle downhill fastest). The full relaxation solves to convergence in a single GPU submit per click, so the isochrone snaps in instantly. Bands draped on the 3D terrain; a pulsing ring sweeps equal travel-time. Backend is static files; everything else runs on the GPU.
The eikonal solver needs a WebGPU-capable browser (Chrome/Edge 113+, or Safari 18+).
The WGSL source ships in the bundle either way — open src/hiking.ts to read it.
Click or drag the terrain to move the trailhead (live) · right-drag to orbit · scroll to zoom · Estenstadmarka DEM (Terrarium z13) · 3.5× vertical & 5× slope exaggeration for legibility.
An unsteady 3D Navier-Stokes solver over a city (Stam "stable-fluids" + a Smagorinsky LES eddy viscosity). Each frame: semi-Lagrangian advection → turbulent diffusion νt = (Cs·Δ)²|S| → pressure projection ∇²φ = ∇·u (red-black SOR, no-flow on buildings). The momentum advection produces real wakes, recirculation and vortex shedding behind the towers — it runs live, because turbulence is unsteady. 70k particles trace the flow in 3D, coloured by speed — you can see the eddies peel off the building corners.
The LES solver needs a WebGPU-capable browser (Chrome/Edge 113+, or Safari 18+).
The WGSL source ships in the bundle either way — open src/wind.ts to read it.
Drag to orbit · scroll to zoom · set wind direction & speed · procedural city · unsteady 3D Navier-Stokes + Smagorinsky LES, stepped live on the GPU.
A 3D globe of the Norwegian energy system — offshore platforms, onshore substations and glowing transmission lines over the real coastline. Runs token-free: bundled public-domain imagery, no Ion account, no external calls. Lazy-loaded so it never slows the first paint.
Loads the CesiumJS engine on demand (~heavy bundle, kept out of initial load).
Drag to orbit · scroll to zoom · nodes pulse in real time.
Every demo above maps to a commercial problem. Each card links to the live scene it describes.
Problem. Grid operators hold millions of asset, sensor and meter points. Standard web maps choke past a few thousand markers — the browser stutters, decisions wait on a spinner.
Approach. Geometry is uploaded once as binary GPU buffers and drawn in a single deck.gl draw call. Pan, zoom and animation touch only GPU uniforms — the main thread stays free.
Problem. Visualizing transfer and congestion across a grid means animating huge particle counts. Done on the CPU it pins a core and drops frames.
Approach. A WGSL compute shader advects the entire particle system on the GPU; the CPU writes a single uniform per frame. Trails accumulate in a ping-pong texture for an instantly-readable flow map.
Problem. Exposure and risk shift constantly. Recomputing and re-tessellating a 3D surface on the CPU each tick is far too slow for live decisioning.
Approach. A custom GLSL vertex shader displaces a 73k-triangle grid from the risk field directly on the GPU; the fragment shader handles colour ramp and lighting. Zero CPU geometry work per frame.
Problem. 3D-globe engines are heavy. Bolted onto a landing page they wreck first-paint — and managed terrain services add per-seat token costs.
Approach. CesiumJS is lazy-loaded only on demand, so the initial bundle stays small. The twin runs token-free on bundled public-domain imagery — no Ion account, no external calls. A free Ion token drops in photoreal terrain when a client needs it.
I'm Bjørn Reppen, a graphics & spatial-data engineer working independently through Betatech AS. I specialise in the hard part of enterprise dataviz: making millions of records render and respond at 60 FPS by pushing the work onto the GPU with custom WebGPU/WGSL and GLSL shaders.
● Available for engagements · fully remote, working globally.
Start a conversation →