How This Was Built

One might expect crazy projects like this to come together on dark winter days, but instead in the hot summer of 2026 I got the idea: what is possible with AI? Why not create a playground for myself, for the fun of it? With content I already have so it's not too much work?

Because starting this from scratch would have obviously been almost impossible. The compulsive side of my personality has been keeping track of a lot of parts of my life: jogging data, concerts I visited, vinyl albums I bought. The creative side of my personality has written a lot: six books, LinkedIn newsletters, a travel blog and more. So the data was there to build something.

The brief to ChatGPT and Claude Design was simple. Come up with visual ideas for a website built around data about my Work (my CV), Music, Tech, Travel, Football and Running. Those were my six rooms. After coming up with several designs (ChatGPT won with a punk-magazine-style proposal), I asked it to come up with a simple HTML version. And in Claude I had a longer conversation about feasible content ideas: how to rearrange the data already there in a meaningful, visible manner? This resulted in a list of ideas, 80% of which have been realized.

The rest of the ideas you see now came from ChatGPT. Because as I have learned in the past three years, it makes sense to have proposals from one LLM (Large Language Model) be judged by another one. Claude's ideas were fed to ChatGPT, which came up with some better ideas. That is for instance how The Predictions Ledger came about, the idea to judge my predictions from my innovation newsletter. The trick, of course, is to have that judgment done by an independent party… an LLM!

Anyway, the real building was done with Claude Code, the subscription version of it, in a couple of evenings in my spare time. From the HTML design from ChatGPT, it had a good jumping point to start. For the different rooms, I created so-called markdown files (basic text files, well-structured for LLMs) to track progress. Then I started to build room by room, with one of the guiding ideas that elements should be re-used. So timeline graphics from the concert lists were also used for my jogging data, as an example.

The trickier stuff was done by Python scripts, written and executed by Claude Code. Yes, that takes a leap of faith, to give AI access to your computer. But that enabled the album wall, or the concert tracker. And all of these things take way more technological and programming knowledge than I have. To put it simply: this site would have been completely impossible for me.

There is no database behind the site, unlike, for example, WordPress. I kept it as simple as possible: some things (concert tracker) take weekly manual updates, some (jogging, album list, concert list) need monthly updates. And as this is a playground, be prepared for more experiments, but as of now (August 2026), I wanted to get this first massive part online and out there.