A huge number of municipalities all share the same tech stack: Granicus/Legistar. You can pull the agendas and minutes of all their board meetings probably going back a decade. From captioning information you can Whisper-transcribe and attribute transcripts of the meetings themselves.
During our last election cycle, I did this for all our board meetings going back to the mid-aughts, using 'simonw's LLM tool to pass each agenda item to GPT 4o to classify them into topical buckets ("safety", "racial equity", "pensions", &c), tying them back to votes, and then doing a time breakdown of the topics (political opponents were claiming our board, which I support, was spending too much time on frivolous stuff).
That's a pretty silly use case, but also a data-intensive one; the things you'd actually want to do across municipalities are much simpler.
You could probably have Claude one-shot a municipal meetings notification service for you.
Granicus is six providers in a trench coat it turns out. IQM2, NovusAgenda, Legistar, Granicus, PriveGov, and CivicClerk are all Granicus projects that share absolutely 0 apis that I've found, and a city having one of these operational is no guarantee they have any of the others.
Legistar and CivicClerk have actual APIs, which is nice, although it's extremely easy for the City Clerk's staff to trip and make the Legistar API unusable.
My experiments with using LLMs to write crawlers for these has been extremely mixed; it's good at getting first page of data and less good at following weird pagination trails or follow-on requests.
All of this led me to build CivicBand (which tracks all the municipalities I can get my hands on) and CivicObserver (which is generalized full-text search alerting for municipalities via email, mastodon, bluesky, and slack webhook)
Yeah, don't get me wrong, they all suck ass, but it's good to know there's one common set of things to scrape to get you lots and lots of cities. Those both sound like very cool projects!
During our last election cycle, I did this for all our board meetings going back to the mid-aughts, using 'simonw's LLM tool to pass each agenda item to GPT 4o to classify them into topical buckets ("safety", "racial equity", "pensions", &c), tying them back to votes, and then doing a time breakdown of the topics (political opponents were claiming our board, which I support, was spending too much time on frivolous stuff).
That's a pretty silly use case, but also a data-intensive one; the things you'd actually want to do across municipalities are much simpler.
You could probably have Claude one-shot a municipal meetings notification service for you.