Ask HN: Is 100% recall search over private data solvable?

[ghita_]

I have been working on AI products for some time, and there is always a RAG / search component to build.

With the plethora of tools out there by now, it's now pretty instinctive to set up with a basic stack.

However, I did see feedback from end users mentioning the search part didn't provide good results. And I also saw several engineering teams going from a standard stack to something way more complex.

Also, there always seems to be edge cases where hybrid search doesn't cut it. (Like asking "search for contracts that DO NOT mention a termination fee" for example).

I'm curious to understand what the community thinks about retrieval systems:

- is it still considered to be a hard problem? - are teams actively trying to solve it, if so how? - will we ever have a system that basically can find any information across any corpus even with minimal indications from users?