Refactored so we are creating the docs DB from the embeddings when they are first needed, rather than loading in from a persistent dir, because this was causing problems.
Moved embeddings from JSON to Pickle to improve read in speed
Deleted old files
Added streaming option to get_llm() - not used anywhere yet, but when used, it allows for streaming LLM output
get_llm()- not used anywhere yet, but when used, it allows for streaming LLM output