Use Azure AI Search Reranker Stand-Alone

Question

Hi all, as I understood Azure AI Search has a built-in reranker, which can be used to rerank retrievals for certain queries. Is it possible to use the reranker API in a stand-alone fashion without setting up a database. I do not want to set up a database as I perform cosine similarity search in memory. I would like to send the query and my retrieved text chunks and rerank them. Is that possible with Azure AI Search? Thank you in advance!

Answer 1

Hi @Arthuro , Azure AI Search does have a built-in reranker, known as the Semantic Ranker. This feature uses language understanding to rerank search results, improving the relevance of the results. However, based on the information available, it appears that the Semantic Ranker is integrated into the Azure AI Search pipeline and works in conjunction with the indexing and query workflows.

The reranking process in Azure AI Search involves feeding a query and results to language understanding models hosted by Microsoft, which then scan for better matches. This process is designed to work with the data indexed in Azure AI Search.

Unfortunately, there is no explicit mention in the documentation about using the reranker API in a standalone fashion without setting up a database. The reranker seems to be tightly coupled with the Azure AI Search service, which involves concepts of search services, indexes, documents, indexers, data sources, skillsets, and synonym maps.

If you’re looking for a solution to rerank your retrieved text chunks based on a query, you might need to consider other machine learning-based reranking solutions that can be used in a standalone manner. These could potentially be integrated into your existing cosine similarity search workflow.