Add support for vector search and document level access control

[sarah-widder]

• adds support for vector search when using Azure OpenAI on your data
• adds support for document level access control when using Azure OpenAI on your data via AAD Auth and the MS Graph API
• resolves endpoint/domain name issues

[avivansh]

@sarah-widder could you please share the resources for adding vector search capabilities.
Currently, https://learn.microsoft.com/en-us/azure/ai-services/openai/reference#completions-extensions, in this documentation vector search isn't been included yet.

Also, how to validate the changes you made in this Pull request associated with vector search are working?

[sarah-widder]

Hi @avivansh we are still finalizing vector search support with associated documentation. We're targeting this in the next week, please stay tuned!

[phfontes]

Hello Sara!

We are trying to implement vector search and we have some doubts. Do I need to make any changes to the.env file? Adding values to the variables you entered in the app.py file, if so, I believe it is necessary to update the readme and .env file.

[hanhsia]

Hi Sarah,
when I send request with vector type information, it always returns 500 Internal server error. Here is my request:
https://.openai.azure.com/openai/deployments/gpt-4/extensions/chat/completions?api-version=2023-06-01-preview
curl -X 'POST'
'https://.openai.azure.com/openai/deployments/gpt-4/extensions/chat/completions?api-version=2023-06-01-preview'
-H 'accept: /'
-H 'Content-Type: application/json'
-d '{"dataSources":[{"type":"AzureCognitiveSearch","parameters":{"endpoint":"https://.search.windows.net","key":"","indexName":"kbindex","semanticConfiguration":"","queryType":"vector","fieldsMapping":{"contentFieldsSeparator":"\n","contentFields":["content"],"filepathField":"file_name","titleField":null,"urlField":null,"vectorFields":["contentVector"]},"inScope":true,
"embeddingEndpoint":"https://.openai.azure.com/","embeddingKey":"","roleInformation":"You are an AI assistant that helps people find information.","filter":"role/any(g:search.in(g, '''1'''))"}}],"messages":[{"role":"user","content":"how to clean c: root?"}],"deployment":"gpt-4","temperature":0,"top_p":1,"max_tokens":800,"stop":null,"stream":false}'
if the queryType is "simple", it works well. I also called azure search api to retrieve data using vectors. It also works.
Is there anything missed in my request?
Thanks!

[sarah-widder]

Hi @avivansh , @phfontes , @hanhsia please see the updated documentation with vector search: https://learn.microsoft.com/en-us/azure/ai-services/openai/reference#completions-extensions
Accepted queryType values are vector, vectorSimpleHybrid, vectorSemanticHybrid. For the embeddingEndpoint you need the full embeddings endpoint, e.g. "https://{AOAI resource name}.openai.azure.com/openai/deployments/{Ada deployment name}/embeddings?api-version=2023-03-15-preview". Hope that helps :)