Rag Pipeline and some other thoughts

[SebastianStehle]

Hi,

do you have any ideas for a RAG pipeline? There are some steps that could be unified, e.g.

• Create a summary of the result before it is added to the context.
• Rerank the results using external APIs like Cohere.
• Rerank the results using another promp.
• Calculate the embedding could just be one step.
• Filter the results based on permissions (if not supported by the vector store).

[sophialagerkranspandey]

If you're trying to build a RAG, we recommend using AzureAI search. You can learn more here: https://learn.microsoft.com/en-us/azure/search/retrieval-augmented-generation-overview

[SebastianStehle]

I think even azure ai search does not cover everything and can be combined with rerankers.