Retrieve linked doc with images in a multimodal RAG app

Hello, I need to retrieve documents using a "join" property.
Let's explain:
I have 2 kind of documents in my index : "test"
images and chunks, they are formatted like this

{
  "type":"image" or "chunk",
  "content":"some textual content",
  "imageVector":[the image dense vector],
  "textVector":[the chunk dense vector],
  "path":"a textual path",
  "decription":"a textual description",
  "joinKey":"a unique id that some images shared with some chunks"
}

Warn : They don't have a parent-child relation.

I want to make an hybrid search (syntax and vector search) on my images (so filtering on the "type" field ="image") :

  • the syntax part of this search as to match the "path" field (with a boost :0.8) and some keywords in the "description" field (with a boost:0.2).
  • the vectorsearch part is a knn search based on the "imageVector" field.

I want to weight the syntax search part to 0.3 and the knn part to 0.7.

BUT I want the result of this hybrid search is not images themselves but the associated chunks through the 'joinKey' field they have in common.

Is there a way to achieve that in a single one search?
Thank you for your help

(PS : I use javascript (or can test in the kibana console)

BUT I want the result of this hybrid search is not images themselves but the associated chunks through the 'joinKey' field they have in common.

What are the chunks? Are they literally the piece of text that is the chunk of text that was encoded?

Or is it the entire document that is "many chunks"?