We are trying to find a solution to indexing and retrieving documents as vectors.
The idea is to index the paragraphs of the document as a child and the document to be the parent document (join solution).
Is there another solution that can prevent the use of parent-child?
In the end our goal is to retrieve a document based on a search text as vector (KNN/ANN) that matches one of the paragraphs but returns the original document.
Parent-Child (Join) Example -
Parent Document: "paragraph1 paragraph2 paragraph3"
Child 1: "paragraph1"
Child 2: "paragraph2"
Child 3: "paragraph3"
All the child include the parent id and indexed as a vector.