A window on the world
What I want this to do for you
Underneath the maps and the checking there is one hope: that Mundu is worth an evening. That it entertains you, and hands you back a little more curiosity than you arrived with.
The Earth is the one subject nobody needs to be sold on. Everybody has a place they want to look at — where they're from, where they've been, the one they've only read about. Put a few thousand years of history on top of that and something changes: you stop being told that a border moved and you watch it move, at whatever speed you like. What you notice then is usually the thing no book bothered to write down — how long a siege really lasted, how small a kingdom really was, how fast an empire can come apart.
That is the whole idea. A window, left open. There are 13 stories to watch so far, and a prompt you can paste into any language model to get one of your own back as a link.
The day job
I'm an AI engineer, and I enjoy it. My work is making business processes more efficient, and for the last few years that has meant integrating AI-powered systems into them. Day to day that means Python, git pipelines and Docker stacks: taking a lot of components — some of them mine, most of them open source — and assembling them into complete applications that run end to end.
Mundu is an outlet, on something profoundly different: no spec, no deadline, and imagination and intuition with somewhere to go.
What I'm judging
The three things I don't hand over
A model writes the code and drafts the stories. It does that far faster than I could, and it is confidently wrong in ways that look finished — code that runs, produces a file and draws a map is not the same as code that is right. So the job isn't typing, and it isn't prompting either: it's deciding what "correct" means here, insisting on a way to measure it, then reading the result assuming something in it is wrong. Three things get read that way.
Artistic direction. The globe is almost black, and that is a decision rather than a theme: everything else on screen is information — a front line, a marker, a route — and information needs somewhere quiet to sit. No labels, no graticule, no country borders. If a pixel is not carrying information, it should be dark.
Historical accuracy. The curated stories are written by a language model too: the research, the dates, the sentence under every marker. What makes them curated is that a human argued with them and refused to publish until they held up. Every story is checked against a table of places and dates decided in advance — not "does it look right" but was this town held by this side on this day — and the check fails loudly. Every marker links to the article it came from, so you can go and disagree with it.
What I can't promise is that it is all right. A single story is thousands of dates, places and borders drawn from open sources, and total precision is not something one person checking a map can guarantee. I do everything I can, and some of it will still be wrong. That is what the feedback form is for — it is the correction mechanism, not a formality.
The feel of it. Fluid and obvious, and nothing else. A globe that stutters while you drag it stops feeling like an object and becomes a picture of one, so nothing off screen is ever drawn. And there is nothing to learn: you spin the Earth with a finger and press play. If something needs explaining, that is a bug in the design — not a gap in the documentation.
Anyway — I hope you like it
It's a hobby, made in the evenings, and it is better than it was last month. Go and watch something happen.