with hours and hours of audio to process, Meridian is on the back of a lot of LLM work. Initially I wanted a tool to provide trend data and the newsletter is a tangible way for me to justify the work. Particularly because it provides value to people without having to do the work.

yesterday I simplified something that should make back search a lot easier: deepgram diarizied data -> split by speaker with metadata -> vectorized -> into pgvector. I should now be able to do give me references Cosmos within this date range, and who did the mentions etc