Leaning towards making the table of contents into db items and throwing the whole notes in a vector db.
So the table of contents db will track when you studied what and to make questions for each topic it'll do a vector dB query to grab the relevant stuff.
Idk it'd be cleaner if it was all postgres but I think i'd need more preprocessinf
OK so I have it make a note with the parent topic and a completed flag because it was just split on pages and now of the last note wasn't completed then it's passed back in.
Will throw this all in postgres and track performance/timing for Anki spacex repetition
I abandoned this nonsense (except the completed bit, that was good) in favor of in-order grouping heuristic. Related notes are going to be next to eachother, when I query, I'll essentially sort by note number and grap like 20 to pass in as context to the llm when it writes an essay prompt. And also factor in the user difficulty/study state stuff from anki
