91awebsi

ouch, and also thank you for calling out the busted.

An LLM titles each theme, but otherwise this is all derived by checking wikidata entities that are 'most read' (according to qrank), then choosing from a set of sparql queries I put together (members of group X, places within Y, etc), for what produces the most results, and then string/substring/bm25 matching across all the streets in the country.

I think the trash smell is too generously allowing an association both in my sparql queries (what to search for), and whats considered a match :/

I wonder too if explaining how a street matched (following the wikidata graph) would make the nonsensical less offending.

I keep meaning to write a post for how this has worked for my projects, I think its been helpful.

For active projects I setup a weekly-ish 'code audit' agent: it runs treepeat to find opportunities to refactor and bumps up enforced coverage numbers. Over time for some projects I lower the '--similarity' threshold to more broadly find similar structured code (potentially more ambitious refactoring)

ooh I'll give yours a shot - thanks for calling my project out on your readme!

I was fortunate to have a contributor push me to consider optimizations a few months back, and it definitely helped improve the speed/memory metrics.

I see the CLI knobs are somewhat similar ('min lines', 'thresholds'), however your 'search' is fascinating. I could see myself wondering "how much error handling is duplicated for HTTP responses?" and your tool would give me a start at an answer. Nice!

Yes, a little AI on the main readme. I started the bones of the thing, but Claude has been behind the keyboard for much of the implementation.