Work
Selected work
Systems I have designed and built, in teams and on my own. Some of it was for employers, so those projects stay unnamed.
B2B platform, Copenhagen
From replaced documents to an event journal
The problem. One monolith database held every document, and each change replaced the whole document, often a huge one. Its history was lost, and every service that needed a document had to ask the monolith.
What we did. We made it event based. Every change became an event that is kept, and the current state is built from the events. A streaming pipeline on clustered services carries the events, built for backpressure, replay and safe schema changes. Then we migrated the historic documents, billions of rows from PostgreSQL, while new ones kept arriving.
What changed. No downtime during the move, and the full history of every document. The events fed the data lake, so the data team got far more detail to work with. The frontend could react to events as they arrived instead of replacing whole documents, which made more interactive tools possible. And services no longer waited on the monolith.
StaChesstics, my own company
Predicting chess tournaments
What I did. I designed and built the data and AI platform behind StaChesstics: predicting tournament outcomes and each player's chances to win, with expected scores and tournament simulations, for players, fans and teams.
What changed. Each forecast comes from simulating the tournament 10 million times (the Monte Carlo method), and the forecasts update as the tournament goes on, so the probabilities and stats are as accurate as they can be.
stachesstics.com ↗A global delivery network
Video on demand and live ingest
What we did. We built the video-on-demand server and an accelerated ingest server for live streams, on a delivery network used worldwide.
A three-person AI startup
One platform to train and serve models
What we did. With two co-founders, we built a general platform for training, versioning and serving models. We set the service boundaries, built the core backend services and moved model delivery to containers.
What changed. It worked end to end. We did not find customers, and stopped.
Facing something similar in your own system?
Book a 30-minute intro call