After the last post How to setup an Azure OpenAI service I explained how the Azure OpenAI service can be set up relatively quickly and easily, I am now focusing on the enrichment of private sources. The goal is to show you how you can make business-relevant data available to your employees without the data leaving the country location or being made publicly accessible.
Today, the focus is relatively simple on PDF files as sources, which are made available to the Azure OpenAI service. The setup and configuration of the service looks as follows:
- Log in to the Azure Portal
- Go to AI Search
- Create a new AI Search service
Important: At least Basic SKU is required - To store data, in our example PDFs create an dedicated Azure Storage account, within a storage container
- Log in to the Azure AI Studio
- Switch to Project > Playground > Add your data > Add a data source
- Add the data source as follows:
- The data source is now created by Microsoft
- Now the documents can be stored in the storage account and the crawl takes place every hour
The prerequisites are now in place with the Azure Search and Azure Storage account. Next, the documents can be uploaded in PDF format, which are then indexed by Azure Search depending on the interval.
In order for the chatbot to access the data, the Azure Search index must be made known to the UI.
The following settings are required for this:
- To do this, open the Azure Web App > Settings > Configuration
- The following parameters must be adjusted:
AZURE_SEARCH_INDEX: [Azure AI Search Index Name]AZURE_SEARCH_KEY: [Azure AI Search Key]AZURE_SEARCH_SERVICE: [Azure AI Search name]
Questions about private resource data can now be asked to the chatbot. If something cannot be answered using the private data, the following message appears.
A successful result could look like this including a reference to the corresponding document.