Happily in practice it has turned out to be great, and the combined space operates really effectively as a community ‘third space’. Obviously because of the private sector aspect of this there’s a significant risk in terms of how sustainable it is, but moderated somewhat by the council owning the actual building both live in.

Mostly this week I’ve been wrestling with a better solution to filtering openrouter providers because a common failure mode is getting stuck on one which is either very slow, or doesn’t actually meet the spec it promises in terms of tool calling.

Edited to add - or just a really good real life one, it would be awesome if everyone had access to a librarian or bookseller with like an hour to sit and talk with you about books and then help you find some.

Oh, and it definitely does not pat itself on the back for suggesting books for you.

(You can see the profile it makes and correct it - it did a pretty good job on me)

It can suggest them to buy, or as audiobooks, or it can find ones currently available to borrow from your local library!

It is at https://leafle.nanosheep.net if you’d like to give it a go :)

It’s pretty fun! It might even work on your local library (I got carried away and spent a while making it work on a few thousand round the world - you can see the coverage here: https://leafle.nanosheep.net/coverage).

https://leafle.nanosheep.net

Alas, I was stymied by not having any cash to work on it, and the unit economics were not very VC friendly (at least I assume that’s one of the reasons why I didn’t get any traction from VCs).

This week I’m thinking about whether it makes sense to provide a location history ‘vault’, designed to let users expose their location history to LLM’s as context.

I’ve spent a while understanding what sort of market would make it viable. I think it does actually work if you can square: 10K participants per major metro area, revenue of about 2.9M per metro area (so say, 5K monthly recurring with about 50 customers).

At that point you could pay data union participants about $5 a week to share their location data with you.

From talking to some previous data union folks, the major challenges are paying out (my target is much higher than any union managed), and people dropping out over time.

My bet is that these are both solvable things by selling data products rather than just bundles of data, and the data source being very passive.

I’m also interested in the idea that such a union should act more like a union than previous efforts in this space, by actively defending members’ data from brokers.

What I’m trying to understand is whether it is viable to pay people ~$5 per week for sharing their location data and demographics based on a 90% share of revenue from sales of data products built on that data. (But without ever selling or exposing individual level data).

More or less my initial approach to this is you take a grid, and you show movements/density on that grid. If necessary you coarsen the grid to avoid reidentification of individuals, and ultimately to get a good picture of the population given the biased sample which is the union membership, you need a statistical model on top which also helps from a privacy perspective.

State actors demanding individual location history is definitely an issue. I have a few possible approaches in mind to defend against that.

The union monetises this by selling privacy preserving aggregates (think ‘where is everyone in London right now’, or ‘where did people commute from?’), and acts on behalf of union members to stop data brokers selling their location data.

Then I switched to cursor and iterated with it on making it work. Once I had the iOS app running, I got cursor to create an android app that looked and functioned the same.

Super interesting experience - took a couple of days of iterating here and there, helped to have them in git repos for when cursor went too far off piste.

Both apps work, they look pretty much like my original picture.. and the code is absolute trash. I don’t know swift, or particularly kotlin, but the code is horrendous. If you wanted to take them any further, you’d rewrite it from scratch.

This was. Rating combinations of Claude 3.7, 4.0 thinking, Gemini 2.5, and chatgpt 4. (Switching between agents when one was getting stuck.)

Observationally, they all had a much harder time with android and rarely produced anything that didn’t need fixing to build.

Great for rapid prototyping in a stack I don’t know, but 100% tech debt.