is it possible to ask a question and parsing all documents?

[be-swarm]

Hi,
I have severals documents(5) embedded whith user information (name, firstname).
I would verify that all documents contains the same name.
my pompt is like "checks that in all documents the names are identical".
Response said yes but relevant source return max 3 sources .
Is it possible to force memory kernel to analyse all sources or x sources ?
I have tested with filters by tag or by document without success.
thanks

[kbeaugrand]

Hi,

Unfortunately, you cannot ask LLM to do that using this RAG strategy. The RAG is able only to read from chunk and get an answer about your question.
For instance (this is not exactly true), if you have a document that told: "documents doesn't have the same name", then it could say to you, no the document doesn't have the same name.

Otherwise, you could store only the document name (e.g: document name: toto.txt, ...) in your vector database. then when you ask about "does all document have the same name?", it might see that some documents doesn't have the same name and answer you no...

The best for me is to use analytical database and use Semantic Kernel plugin to connect to your analytical database and use query to get distinct document names...
There is no easy way to do that I think

[be-swarm]

Hi,
Thanks for your response.
when I say name it's the name of a person.
ex: every document contains Thierry ROUSTAN or ROUSTAN thierry or ROUSTAN.
I would like to know if the name is similar in all documents.

[kbeaugrand]

Yes it's the same for me;) you cannot expect a relialable answer.

Instead prefer vectorize separatly the name in an index, and try to verify if they seems to be part of the same cluster (meaning what is the distance between each vector to each other)...