infra . data . engineering . ai
Systems that hold their shape.
Strategic by default, hands-on when it counts, relentless until it works. A decade of building data and ML platforms — research pipelines, production ML, and now LLM infrastructure as CDO at Swoep.
The register
Essays
Things I have learned. Things I built, broke, and eventually figured out.
Price fencing: the fence that sorts your customers for you
One price is a compromise nobody asked for: too high for the student, a gift to the procurement department, and usually both mistakes at once. A fence — an objective condition that costs the customer something — lets people sort themselves into price brackets without you ever guessing who is who. Here is what a single price actually costs, why the revenue curve is a lazy hill that hides your mistake, how much each extra fence is worth, what fences are made of, and how leaky a fence can get before it stops earning its keep.
Van Westendorp: four blunt questions and one elegant X
Ask someone what a thing should cost and you get a shrug. Ask where the edges of acceptable are and they answer with startling consistency — which is the whole trick behind the price sensitivity meter, a method from 1976 that survives on four survey questions and no statistics at all. Here's how forty-eight scattered answers become four smooth curves, why two of them run the other way, where the famous smoothness actually comes from, and why the point labelled 'optimal' is the most dangerous number on the chart.
How to solve the 2D rectangle variant of the IQ fit puzzle: math and code
Twelve ball-pieces, an 11×5 board of fifty-five holes, and one exact covering. Symmetry cuts the search space for free via the dihedral group, a counting argument proves that a full board is a solved board, and a Rust solver packs the whole grid into a single u64 so every placement is a bitwise AND.
Playground
Experiments
Small things, built for the joy of it. The live ones run right here.
Live demo · UUID v7 inspector
A timestamp in disguise.
A UUID v7 starts with 48 bits of Unix-milliseconds. Generate one and see the timestamp hiding inside.
Diffie–Hellman playground
Watch two parties agree on a secret while you see every byte they exchange — then switch sides and try to break it as Eve.
Live demoGradient descent playground
Roll downhill on a loss surface: dial the learning rate up until the steps overshoot and diverge, and watch the trajectory pick its way across a bowl, a ravine, or a saddle.
Career
Experience
From bedside medicine to terabyte-scale genomics to production ML platforms to AI infrastructure. Each layer still carries the ones beneath it.
2025 — now
Chief Data Officer · Swoep
Building the AI infrastructure from the ground up: LLM agents, MCP servers, CI/CD for LLM applications, and the platform migrations that keep it all fast and affordable. Together with the information security officer, I set up and ran the ISMS that took us to a successful ISO 27001 certification.
2019 — 2025
Senior Data/ML Engineer & System Architect · Gemvision
End-to-end ownership of the data platform and infrastructure: an MLOps platform for ML model development and deployment, ETL pipelines processing 1M+ records a year, zero-downtime deployments on AKS, and ISO 27001 + NEN 7510 certification done the hard way.
About
Systems, layer by layer.
I started out in medicine. A decade ago I traded the bedside for data: a PhD in computational genomics at Erasmus University that meant building pipelines to chew through multi-terabyte datasets. Research taught me rigour; the pipelines taught me engineering — and I never looked back.
Read the full storyLet’s talk shop.
Build something that lasts.
Data & AI platform architecture, LLM infrastructure, MCP and agent work — or just a good conversation about any of it.
More ways to connect