Multivector/Multimodal Vector Search in Individual Queries

[txhno]

Consider a Product Search Platform where each "product" (or point, in Qdrant terms) includes both image and text data. For example, a product may have multiple images (e.g., 5 images) and a text attribute (e.g., title).

I aim to store embeddings for all 5 images (multivector) and the title (multimodal embedding) within the same point. I also want to query these multiple vectors simultaneously. For instance, if I query using one image embedding and one text embedding at the same time, the goal would be to find the point with the closest matching image and text embeddings.

I am still working out the exact logic for this. If you have any suggestions or insights on how to implement this proof of concept (or if it's a viable approach), please share your thoughts!