is it capable to answer for openend question?

[gokalpcaymaz]

lets assume that I have a document which has the review contents of one products. is it possible to get answer for "regarding to review comments, give top 5 concern about the product" or "regarding to review comments, suggest improvement point of this product?" so these questions are not directly addressed into file search. So I m wondering that this architecture is capable for this type of questions.

[pamelafox]

It can answer the questions as long as it has enough information to answer it. By default, we only retrieve the top 3 matching document chunks for a particular query, but you could extend that to retrieve the top 5, 10, or even more, though you need to limit to your model's context window.

If you are always dealing with a structured document, like reviews, you might also want to consider RAG on database rows, like this repo:
https://github.com/Azure-Samples/rag-postgres-openai-python/

This repo is optimal for RAG on unstructured documents with different formats and length.