Sure, it’s pretty dumb/naive implementation, would need to be a lot more efficient to scale. Basically had Claude write polite/low touch dumb crawlers for a bunch of local sites (library, local events spaces, luma, theaters, maker spaces) and whip up a little frontend to let our family and friends manage a little text description of what their family members like, constraints, that sort of thing. Once a week, the crawlers look for new events, add to a db, and then run through the list and ask our local LLM to grade each event, given the text description and constraints. Take the top 20ish for the following two weekends and email out. It’s been super helpful, lowers the activation energy to go to more local events.

▲

saturn8601 2 hours ago | parent [-]

Do you really need a powerful Mac to do this? Wouldnt this work on a base model or even an old Windows computer?

	▲	ericd 5 minutes ago \| parent [-]
		Probably don’t really need it, just makes it easier. I tried it with some potato class models first and they would ignore or mishandle vaguely worded constraints. Sometimes it would try to send us to events that were obviously in the middle of the school day, and I’d look at their reasoning traces and there’d be some really boneheaded mistakes in there. If it reliably sends you a decent fraction of garbage reccs, the emails stop getting opened, my SO has no patience for that.