I thought it would be a nice 30-minute Sunday morning activity to update this little API to use the new 2025 IMD data.

The new geography data is also much higher resolution (nice!), but that means the SQLite file is now 500MB, so this little hack has broken (boo!)

I have other things I need to do

Unfortunately, "vibe electronics" doesn't appear to be an option yet. Like humans, chatgpt struggles with how many fingers a transistor has.

I think a career doing Test Driven Development is unhelpful when you're doing electronics. Creating a failing test has cost me £60.

It feels weird, but is it bad? The reduction of external dependencies and build complexity feels like it's worth it?

Committing a 65MB database file to version control still feels weird though.

Appreciated Mohamed Hassan's talk at @brightonruby.com and @simonwillison.net's datasette work for getting me to take SQLite more seriously.

Talk: brightonruby.com/2024/litesta...
Also <3 Simon's excellent work in the open that helped me debug my spatialite docker img: github.com/simonw/datas...

I've repeatedly come across the need to find intersecting polygons for a location from a read-only source, but usually end knee-deep in PostGIS. Once I got it working, SQLLite with Spacialite is a lot more suited for this. The whole thing contained in a read-only docker image.

www.londoncentric.media/p/drag-queen...

I think it’s a very good view of your work, but more of a wistful view of a younger, more hopeful self.

But a key thing to note: no matter how prodigious you are in your personal data production, it’s not close enough to being useful for training or tweaking a generative model with. You’ll be looking at prompt augmentation.

I have plonked my Apple reminders into my morning journal prompts, but haven’t had much of an effect yet. Needs more tweaking maybe: github.com/james/ducky/...