Everything in RAG rests on one idea: embeddings turn text into vectors where similar meanings sit close together. That is the whole trick - you can now find text by meaning rather than by keyword, because "how do I cancel" and "end my subscription" land near each other in vector space even with no shared words.

Get this intuition solid, because retrieval quality, and therefore your whole system's quality, depends on it. No maths required to use it well, but the mental picture of meaning-as-position is non-negotiable.