Can you explain how the database and its documents are searched for this query information?

@@jannik93 from the document chucks there is embeddings created and stored in the vector database. When you ask a question this question is also converted as embedding with that the cevtor database is searched for matching results with distance value and returns the text from the documents from the database and passed to the LLM with question and answer and how the LLM should handle it.

@@thedevmachine thank you! For understanding: You mean "document chunks"? Are the embeddings created by the retrieval model or the ai?

@@jannik93 you read the document as text create chunks from it then from each chunk you create embeddings and store it on the vector database. To use to ask a question that question is also conveted to embedding so you do search and you get result or results depending on your question the result contains the relevant chunks as text and you pass the original question and the answers with a prompt to a LLM to generate a natural response.

@@thedevmachine I think I got it now. Thank you very much!

I mentioned those aspects. I didn't cover chunking because of the issue of timing in a short (60 seconds). Chunking is a pre-step in RAG systems and not an essential one. Some documents don't require chunking, such as Tweets for example. I did mention vector databases.