support for multiple vector embedding in one document

see we all need this feature to search with multiple vector embedding,
let’s discuss more
let we have a social media database in which posts have description embedding as well as image embedding now when we search with a search term we should match it to the image embedding as well as description embedding and sometimes comments embeddings so we should search one embedding for all the fields at once. so that we can query it faster.

One approach that will work here is of creating two separate vector indexes on the two different fields, performing a query on both, and then finding an amalgamation between the two result sets

Now, essentially, what you are doing is executing two vector search queries in parallel (against the two separate vector search indexes) and combining them using an algo like RRF.
The query processing pipeline involves the following steps:

  1. Query Execution: For each query, the system executes the query against the indexes to produce a ranked result set.

  2. Result Merging: Use the Reciprocal Rank Fusion (RRF) algorithm to merge the rankings from multiple result sets into a single, unified result set. The RRF algorithm combines the reciprocal ranks of the same document from different result sets to compute a combined score. The combined score is then used to rank the documents in the final result set. The formula used in the RRF algorithm is as follows:

RRF Score=1 / (k+ doc position in the result + penalty) You can read more about it here

  1. Query Response: The unified result set is then returned as the final response to the user’s query.

Let me know if this makes sense to you and we can discuss further about your use case?