The simplest approach to deploying a chatbot is to do so without any coding. This is precisely what I aim to demonstrate. Such a chatbot can be tailored to utilize your custom documentation, knowledge articles, or any other resources you wish to integrate. This method simplifies the process, making it accessible even if you’re not well-versed in programming. By leveraging existing documents and knowledge bases, you can create a chatbot that is both informative and aligned with your specific needs and content.

Cpncept

The deployment involves setting up a dedicated web page featuring a chat interface. This interface interacts with Azure AI Search, fetching content from your custom documents. The interface then seamlessly combines this content with user queries before forwarding them to Azure OpenAI services.

Setting Up the Chatbot Co-Pilot

Prerequisites
To initiate our first Chatbot Co-Pilot, certain prerequisites are necessary:
– Azure Blob Storage Account: For storing the documents.
– Azure AI Search Instance: To enable content retrieval and indexing.
– Azure Open AI Account: For storing the documents.

To setup the Azure AI Search follow the following steps:

• Open the Azure Portal
• Search in the Search bar for Azure AI Search
• Click + Create

• Select the Subscription, Ressource group and enter a name and a location
• Click Review + create

Preperation

Now we have to do some preparation steps. For this open the Azure AI Search service and select an Semantic ranker plan:

• After this we can store our documents into an blob account. But bevor we can upload them we need an container:

• Once the container is created you can upload the documents into this container

Setting up the chatbot

Open the Azure Open AI Studio and create a deployment of an gpt-35-turbo and text-embedding-ada

• Go to the Chat section
• The first step is to define a system message to instruct the chatbot who he are and what he should do. My example was (“You are an senior expert who can answer all questions about intune. Please only answer questions related to intune and only answer if you are sure.”). Once this is done click on Apply changes
• I changed the temperature to 0 to make sure that we have as less as possible hallucination

• Change the section to “Add your data” and click on + Add a data source

Now we have to configure the connection to the Blob and the Azure AI search. To do this select Azure Blob Storage as data source. After this you have to select the subscription where you blob and your AI search is located. After this you have to enter an name for the index and select an schedule. In my case I do not update the documents regularly for this it is sufficient to create the index once. To get better results we also activate the Add vector search to this search resource flag and select our embedding model. When all this configurations are done you can click Next.

In the next window we have to check the I acknowledge that using sementic search will incur usage to ma Azure AI Search account and we have to select the chunk size. This is the process of breaking down large pieces of text into smaller segments. I selected here the default settings and checked the I acknowledge that adding vector embeddings will incur usage to my account. Click Next

• Click Save and Close. The configuration can took some seconds.

Now the integration from the search is also setuped. The only thing what we now have to do is to deploy the webpage. Also here we have an very nice wizard which guides us through the whole process of the deployment.

Click here on Deploy to and select “A new web app…“.

Now we have select the subscription and ressource group where the app service should deployed to and we have to enter an name for the app. This name will late also be an part of the url. When you use Free as pricing plan you have no costs bur also no slas for the web app. Click then on Deploy.

The deployment will take some minutes. Once the deployment is done you can open and test your bot.

This is the default look and feel of the bot. You can already use it to ask questions about your documents.

But lets have also an quick look how you can customize the look and feel. For this you can clone this repository: https://github.com/microsoft/sample-app-aoai-chatGPT and make the changes you want to do.