How can I just ask AI a broad question over large data? Is RAG the answer?

So recently we've been experimenting with semantic indexing in our product. So far it's going great, such a powerful tool.

For context, we've been doing tests with codebases (obvious, I know).

So far we've been able to ask questions like: "How do I fix this bug in the user's list of generated websites?" or "Where do I have to go to make the generated ads better?" in which the query gives the tool something concrete to look for.

But what if I'd like to ask something broad like: "Where are the the biggest performance bottlenecks" or "Tell me if there any obvious bugs in the code", when there's nothing really specific to compare agains't the vector embeddings?

Any known ways on how people are solving this?