Can we use Azure DevOps Wiki as a knowledge base for the chat bot?

Question

Hello!
We have a huge Knowledge Base for our internal applications which contains a lot of documentation, policies and guidelines, tutorials, question answers and documents and forms related onboarding and so on which is currently maintained on Azure DevOps Wikis. We want to have a ChatBot which can read and analyze the information from the Knowledge Base and respond to user queries with meaningful answer without letting the user to redirect to the Azure Wikis url and then go through whole KB article or the document.

Question - could the Azure Wikis serve as a knowledge base for a ChatBot? If yes, then what would be the recommended Azure Services I should be learning and using to build the ChatBot?

PS - I have no experience in using ChatGPT like services and I myself is getting started into using AI services.

Answer 1

Arpan Mehetre Greetings & Welcome to Microsoft Q&A forum!

As long as you meet the following requirements you can use Azure OpenAI Service on your data

• A public website, such as Using your data with Azure OpenAI Service - Azure OpenAI | Microsoft Learn. You can't add a URL/Web address with access control, such as ones with a password.
• An HTTPS website.
• The size of content in each URL is smaller than 5 MB.
• The website can be downloaded as one of the supported file types.
• Only one layer of nested links is supported. Only up to 20 links, on the web page will be fetched.

could the Azure Wikis serve as a knowledge base for a ChatBot? If yes, then what would be the recommended Azure Services I should be learning and using to build the ChatBot?

AFAIK, to achieve this with Azure DevOps wiki articles you can extract the relevant information from the articles and store it in a structured format such as a database or a knowledge graph. You could then use this structured data as the input to Azure OpenAI to generate responses to questions.

Azure OpenAI On Your Data supports the following file types:

• .txt
• .md
• .html
• .docx
• .pptx
• .pdf

You can convert data from an unsupported format into a supported format and optimize the quality of the model response by ensuring the conversion:

• Doesn't lead to significant data loss.
• Doesn't add unexpected noise to your data.

If your files have special formatting, such as tables and columns, or bullet points, prepare your data with the data preparation script available on GitHub.

For documents and datasets with long text, you should use the available data preparation script. The script chunks data so that the model's responses are more accurate. This script also supports scanned PDF files and images.

You can also check the Chat with your Azure DevOps data and What is custom question answering? for more details.

Do let me know if that helps or have any other queries.

Answer 2

Hi @Arpan Mahatra ,

Yes you can have a website hosted in Azure App Service with Entra ID based authorisation, and you can use RAG pattern to ingest the wiki content into a backend solution. Easier and straight forward would be to upload content to a blob container, which can be indexed by Azure AI Search to return content matching with the queries/questions.

Azure Open AI can integrate from Web app and retrieve responses from the AI Search.

Answer 3

Copilot studio now supports AzureDevops as a "connection type" which should, in theory, allow the graph search capability to kick in.

However, I have not had any luck getting it to actually work.