use AISearch import and vectorize data wizard with azure-search-openai-demo

[evan2k]

I was trying to use Azure AI Search import and vectorize data wizard (from the portal) with the app.
It seems that the wizard is quite opinionated on the name of index fields it uses when
creating the index, which makes it incompatible with the app, unless you make changes to
the code and use the same names. Am I missing something or am I correct in my assumption?
thank you.

[pamelafox]

I just helped another team use the app with the integrated vectorization. These were the changes they had to make.

1. Modify this code in approach.py to pull out the correct field names.

 
                documents.append(
                    Document(
                        id=document.get("id"),
                        content=document.get("content"),
                        embedding=document.get("embedding"),
                        image_embedding=document.get("imageEmbedding"),
                        category=document.get("category"),
                        sourcepage=document.get("sourcepage"),
                        sourcefile=document.get("sourcefile"),
                        oids=document.get("oids"),
                        groups=document.get("groups"),
                        captions=cast(List[QueryCaptionResult], document.get("@search.captions")),
                        score=document.get("@search.score"),
                        reranker_score=document.get("@search.reranker_score"),
                    )
                )

You probably want id to be "chunk_id", and also for sourcepage and sourcefile to be "chunk_id", as I've been told that field contains the filename. You need filenames for citations to work. The embedding field should be "text_vector", I think. The rest of the fields are optional and don't need updating.

2. Change this line of code in approach.py to "text_vector" (or whatever your embedding field is called) -

return VectorizedQuery(vector=query_vector, k_nearest_neighbors=50, fields="embedding")