-
@ Matt Hamilton
2025-02-25 14:24:45
Pretty good search results using a combination of lexical search and function calling.
I'll continue experimenting with semantic search (vectorized embedding), but this seems to be working pretty okay for right now. It helps that the data is localized to begin with.
Speaking of speed, it takes ~4s but that's before any optimization of model choice, hardware, caching, etc.
(The response is accurate based on the information provided by the tool response, but it's just dummy data for now.)
https://static.noauthority.social/media_attachments/files/114/064/939/214/239/647/original/cff789aef824756b.mp4