Which is more accurate: 1 Assistant with 10 files or 10 Assistants each with 1 file?

API
,
1

Imagine I want to create an accurate RAG assistant.

I have two options:

  • Create one assistant that has 10 files loaded into it that it can retrieve from.
  • Create ten assistants each with 1 file, and have the user’s request routed to the correct one.

Which of these will lead to more accurate responses for the user?

1 Like
2

You should be able to use functions to dial in the 1 assistant and have it pick from the 10 based on 10 functions. I don’t think you need 10 assistants. I’m sure there are other ways to do this with one assistant, though, maybe some other people will weigh in.

1 Like
3

Thanks. But which method would provide more accurate results? I would’ve thought 10 different assistants (10’s just a number by the way, I really just mean multiple vs single file assistants) would be more accurate than one that has all ten files, but I don’t have anything to go by.

4

depends on the type of questions. For example, what if you want to search for patterns across the ten files?

5

True. In this context, each file is categorised separately enough there won’t need to be cross-checking.

Which means, yes, I could simply have an assistant for each category, but I wanted to streamline it into one for a more seamless user experience. But now unsure whether it’s more accurate to have the main assistant take a request and route it to the relevant category assistant or the main assistant itself has all the files to answer from.

6

Assistants are non-deterministic so if you have one that routes to your desired file, it may or may not find the right file

1 Like