Az Function Embedding Service Proposed Updates

[luisquintanilla]

User Flow

graph LR
    UI[User Interface] --> |Uploads document|Backend[Backend Service]
    Backend --> |Uploads document|Storage[Azure Blob Storage]
    Storage -->|Blob trigger| Function[Azure Function]
    Function -->|Generate embeddings and index document| Search[Azure Cognitive Search]

Tasks

• [x] Fix 500 error thrown by api/documents endpoint with handler OnPostDocumentAsync

System.Net.Http.HttpRequestException: TypeError: Failed to fetch ---> TypeError: Failed to fetch Exception_EndOfInnerExceptionStack at System.Net.Http.BrowserHttpInterop.d__13`1[[System.Runtime.InteropServices.JavaScript.JSObject, System.Runtime.InteropServices.JavaScript, Version=8.0.0.0, Culture=neutral, PublicKeyToken=b03f5f7f11d50a3a]].MoveNext() at System.Net.Http.BrowserHttpHandler.CallFetch(HttpRequestMessage , CancellationToken , Nullable`1 ) at System.Net.Http.BrowserHttpHandler.g__Impl|53_0(HttpRequestMessage , CancellationToken , Nullable`1 ) at Microsoft.Extensions.Http.Logging.LoggingHttpMessageHandler.g__Core|5_0(HttpRequestMessage , Boolean , CancellationToken ) at Microsoft.Extensions.Http.Logging.LoggingScopeHttpMessageHandler.g__Core|5_0(HttpRequestMessage , Boolean , CancellationToken ) at System.Net.Http.HttpClient.g__Core|83_0(HttpRequestMessage , HttpCompletionOption , CancellationTokenSource , Boolean , CancellationTokenSource , CancellationToken ) at ClientApp.Services.ApiClient.UploadDocumentsAsync(IReadOnlyList`1 files, Int64 maxAllowedSize)

• [x] Deploy Azure Function to Azure
  • Should have access to all resources in the deployment, specifically OpenAI, Azure (Blob) Storage and Azure Cognitive Search
• [x] Create shared embedding service
  • Should use OpenAI embedding models to generate embeddings.
  • Should use Azure Cognitive Search instance to index documents
• [x] Update AzureSearchEmbedService (to bsaically do what PrepDocs is doing for chunking / embedding generation / indexing)
  • Should use shared embedding service to generate embeddings
  • Update SearchDocument schema to include embeddings
• [x] Update PrepDocs to only upload documents to Blob Storage. Let the Azure Function handle embedding generation and indexing.

[luisquintanilla]

How to (manually) test

1. In the Azure portal, navigate to the storage account tied to the deployment.

2. Use the Storage Browser to upload a PDF file to the content blob container.

   

   This assumes that you used the default azure storage container. Otherwise, use the value of AZURE_STORAGE_CONTAINER set in the .env file.

3. In the Azure portal, navigate to the search service tied to the deployment and click on the Seach explorer

4. In the search explorer, conduct a keyword search for terms that might be found in the document you uploaded. For example, if the documents you uploaded contain healthcare information, you might want to use that as the keywords. If successful, it should return relevant search results.

[igrgurina]

This would be incredibly useful

[luisquintanilla]

Marking this issue as complete since this service now lives in the EmbedFunctions project.