Automated indexing has streamlined document search workflows but semantic relevance and setup complexity still need improvementWe used the semantic search capabilities of Azure AI Search, but we haven't gotten good results in the semantic search. So we are exploring with ChromaDB, and Cosmos is having the capability of doing the semantic search as well. We are exploring that. A few queries we use analytics search, which works and is good. Analytics search is good. We are trying the ML capabilities of the product since we are using Databricks and other tools for building the models, MLflow, and related items. We are still working on proof of concepts, which could be better with ChromaDB or Cosmos or vector search or inbuilt Databricks vector stores. Language processing is not about user intention; it's about the context. If there is a document and you want to know the context of a particular section, then we would use vector search. Instead of traversing through the whole document, while chunking it into the vector, we'll categorize and chunk, and then we'll look only at those chunks to do a semantic search. When comparing Azure AI Search, I'm doing a proof of concept because with ChromaDB I can create instances using LangChain anywhere. For per session, I can create one ChromaDB and can remove it, which is really useful for proof of concepts. Instead of creating an Azure AI Search instance and doing that there, that is one advantage I'm seeing for the proof of concept alone, not for the entire product. I hope it should support all the embedding providers as well. Is there a viewer or tool similar to Storage Explorer? We are basically SQL-centric people, so we used to find Cosmos DB very quick for us when we search something and create indexes. I guess there is some limitation in Azure AI Search. I couldn't remember now, such as querying limitations. I'm not remembering that part.
