Hybrid Search + Semantic Ranking

[PDDhillon]

Hi there. I've implemented a web application using kernel memory with Azure AI Search to index my documents and ask questions on them. I know that hybrid search is possible through a boolean flag in the appsettings.json, however, my question is whether semantic ranking is possible through kernel memory? If not, is something that is coming in the future?

[dluc]

Hello, semantic ranking is not yet possible/we haven't tested it in KM yet. It should be available in the next months after refactoring the code to use the new SK memory classes we're developing. Still early though, we need more testing and there might be extra work.

[PDDhillon]

Thanks for getting back! Looking forward to when this is available

[luismanez]

hey @dluc is there any update on this one? ... we're very interested in this scenario.
Cheers!

[dluc]

We likely won't add Azure AI Search Semantic Ranking in the KM connector, because we want KM to reuse SK connectors once they're ready. SK vector store connectors are currently missing the ability to combine multiple filters with AND and OR.

I would raise the feature request on Semantic Kernel repo. Here's the file that I think will be extended: https://github.com/microsoft/semantic-kernel/blob/main/dotnet/src/Connectors/Connectors.Memory.AzureAISearch/AzureAISearchVectorStoreRecordCollection.cs