Got openai with my data from sharepoint working. Have I overcomplicated thing?

Question

Hey! I have been able to create an azure ai search that retrieves data from a sharepoint folder, following this guide https://learn.microsoft.com/en-us/azure/search/search-howto-index-sharepoint-online. Was able to do this to use the index for chatting with azure openai with my data and got it working. So far awesome.

I got two questions. How do I vectorizes or create a sematic ranking on my index? I feel that the results I'm getting when searching is sub-par and I think it's the index (keyword).

Second questions: Am I overcomplicating things using postman to get my data source and index it? Is there any easier solutions out there?

I don't mind using either Azure AI Studio or Azure OpenAI for this but I will use a gpt as LLM in the end.

To recap: My goal is to deply a webapp using azure (studio or openai) that uses data from sharepoint to chat with. Am I on the right path or not?

Answer 1

@Axel Qvarfordt sounds like you're on the right track but there a few things you can do to improve your search results.

Currently, your Azure AI Search index uses keyword-based search, which might not capture the semantic meaning of your data. Try using semantic ranking in Azure AI Search because it improves search relevance by using language understanding to rerank search results. To use semantic ranking, you need to add a semantic configuration to your search index and add parameters to your query. This configuration establishes a priority order for fields that contribute a title, keywords, and content used in semantic reranking. You can follow the steps provided in the Quickstart: semantic ranking guide to implement this. Note that semantic ranking doesn’t use generative AI or vectors. If you’re looking for vector support and similarity search, you might want to explore Vector search in Azure AI Search.

While Postman is a popular tool for sending REST requests to Azure AI Search, there are alternatives. You can use the Azure portal’s Import data wizard, REST APIs, or Azure SDKs for .NET and Python for indexer-based indexing. If you need a SharePoint content indexing solution in a production environment, consider creating a custom connector with SharePoint Webhooks, calling Microsoft Graph API to export the data to an Azure Blob container, and then use the Azure Blob indexer for incremental indexing.

To address your last question, both Azure AI Studio and Azure OpenAI can be used to deploy your web app. Since your primary goal involves deploying a web app for chatting using a GPT-3 model and your data, Azure OpenAI seems like a better choice due to its easier GPT-3 integration and focus on OpenAI APIs. You can deploy a chat web app that uses your own data with an Azure OpenAI Service model by uploading your local data files to Azure Blob storage and create an Azure AI Search index. Your data source is used to help ground the model with specific data. You can also use the available standalone web app to interact with Azure OpenAI models using a graphical user interface.

Let us know if you have further questions.

-Grace