RAG method can retrieve information from our own knowledge base, but how can we ensure it accurately answers descriptive questions?
E.g. If I store entire text from books (Project Gutenberg), and user asks a descriptive question like “How many books authored by O Henry are available?” then I want the exact number. Descriptive question would be any such question which would effectively require scanning all documents rather than finding documents “similar” to the context of the question.
If this is not possible to achieve with RAG and requires specifically tailored solutions using Solr, etc. where faceting is possible, is it possible to at least infer from the question when to send the query to RAG pipeline and when to use hard index search?