We are using llama index to search the

putting 300k documents into a vector db and using simple top-k retrieval is not going to cut it.

You likely need hybrid search (I don't the milvus supports this) or you need to organize your data into several indexes (by topic? Category?), and use a router, retriever router, or sub-question engine on top of the data

Likely sentence window retrieval might help as well, if the data is easily split by sentence

Of course, all of the requires re-embedding/re-building indexes 😅 Planning for indexes of this size is key

@Logan M thanks. Will try multiple index and retriever router . What is the recommended size for a vector database collection?

It's hard to say tbh, it really depends on your data I think

The better you can sort things into categories, the better I think