Vertex AI Search allow for Chunking

[nilskk]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

Vertex AI Search on langchain only works if extractive answers or extractive segments are turned on. This directly rules out the chunking feature of Vertex AI Search and therefore using Vertex AI Search for RAG applications. Is there the possibility to implement Vertex AI Search to return a specified number of chunks instead of the extractive answer for unstructered documents?

Motivation

The Chunking feature and Layout Parser of Vertex AI Search can't be used with langchain, which limits the possibilites of this service.

Proposal (If applicable)

Add a chunking option to return the top n chunks for a search request