Anyone is able to explain query decomposition properly? What is going on under the surface if I may ask?

Awesome - yeah I don’t blame you guys at all for not focusing on documents right now. I do find it generally more useful to use keyword indexing because it just helps categorize pieces of data as relevant or not relevant - may be a good default relative to the list

Is there a way with llama index of splitting out general background knowledge of the LLM versus the context you are feeding in?