hi.

It's a bit strange, standing here.

I don't consider myself a developer.

…and more.

but for now: Berlin's open data.

We all love data.

Berlin publishes all of it.

So this should be easy.

Where do you find…

Berlin's official open data portal. This should be easy.

Zero results. OK. Wrong word, maybe?

The technical term. Also zero.

37 results! …all of them planning documents.

Six results! You just had to know the magic words.

The dataset page. The actual link hides further down.

Not the data. XML that DESCRIBES the data.

The data! 16,217 polygons of fresh air. In GML.

No blame. But this is
what fragile looks like.

The portal again: zero results for election results.

The election office. It sends you elsewhere.

Excel files. For humans. Behind hash-URLs. Plus a cookie banner.

Berlin does not even host its own election data.

One result: a map IMAGE. No data.

A ZIP from 2021. From VBB. Not Berlin's. Dead end.

The city has nothing.
The community does.

Every tram stop. Mapped by volunteers.

You just need to learn a query language. Obviously.

And there they are. All 960 of them.

Remember the quiz?
Now you know why nobody could answer.

You know me by now.
I got obsessed.

I needed to collect, to clean, to aggregate, to map. And to make it fast.

I bought another domain for a side project I never knew I'd start:

One address. Every layer.

…and 45 more.

Nobody wants to browse 65 layers.
So I aggregated.

How? Step one:

Every raw value becomes 0 to 100.

noise: low → 100 · kita next door → 100 · kita 1,000 m away → 0

Step two: layers → one dimension.

Step three: five dimensions, equal weight.

(29 + 10 + 74 + 63 + 100) / 5

The score sees the city

Noise first. The score joins in.

Two datasets.

Two datasets, one truth.

Where the city protects housing

Housing quality first. Protection joins in.

The map renters actually need.

Mobility & supply

Groceries first. Transit joins in.

There is a Späti around the corner.

Talking about cold beer
from the Späti…

Remember this summer's
40°C dog days?

“noch recht lückenhaft”

taz on the official cool-places map, June 2026

“Wollt Ihr uns verarschen?”

Berliner Kurier, testing the same map, August 2025

I thought: I can do better.

In August, the taz went counting around Gendarmenmarkt.

“Zuflucht vor dem Hitzeschlag”

taz, August 2026 · via hitze.navigator.berlin

*.navigator.berlin
hitze was only the first.

Remember?
I don't consider myself a developer.

I didn't code this.

I coded specced it.

So I just typed
“build me a city atlas”?

Not quite.

Some call this “vibe coding”.

There are no vibes
in a spec.

Agents at work? Hendrik showed you.
My part: the spec.

Slow. Expensive.
Worth it.
tokens and hours in, better software out

Right. This is a Svelte meetup.
Some tech, then.

What did Svelte actually buy me?

And these slides?
Svelte too.
built with Animotion · by Joy of Code

One more bet:
the next users won't be human.

navigator speaks WebMCP.
11 tools, right in your browser.
An emerging standard. Already live.

One more thing.

What if you could tell the map
what you want?

how the Kiez-Finder works

Kiez-Finder

Let's find out. Live.

One more dimension: how people vote.

Now imagine matching this
with actual listings.
(if there were any flats to list)

“Not so good at running them”?
This one's different.