I’d use the LLM as an intake layer, not the database: raw sources stay untouched, extracted claims become structured rows with citations and open questions.
Accessibility data collection, I am running a local application with a mongoDB connection so that when I decide to deploy it fully after validation then I can without having to collect the data again.
Honestly data structures don't have to be perfect unless you are feeding it into another service that you don't control. If you end up wrapping the research into another piece of the puzzle you control both sides anyways.
I’d use the LLM as an intake layer, not the database: raw sources stay untouched, extracted claims become structured rows with citations and open questions.
Accessibility data collection, I am running a local application with a mongoDB connection so that when I decide to deploy it fully after validation then I can without having to collect the data again.
Honestly data structures don't have to be perfect unless you are feeding it into another service that you don't control. If you end up wrapping the research into another piece of the puzzle you control both sides anyways.