Introduction

This is a simple C# demo highlighting how to have conversational interactions with private data in Azure using Azure Cognitive Search and Azure OpenAI.

Azure Cognitive Search is used for Retrieval Augmented Generation.

Requirements

You should have the following resources set up on Azure,

• Azure Storage Container: This will host your private documents (PDF, Word, Excel,...) that you want to use as the knowledge base. The storage account is added as a data source of your search service.
• Azure Cognitive Search Service: The service should be set up in a standard tier which will ensure you can use Semantic Search. Add the storage container as a data source for the search service and set up the indexer, index, and Semantic Search configuration.
• Azure OpenAI: Add a model deployment to your Azure OpenAI service, either GPT 3.5 Turbo or GPT 4.

User Secrets

Add user secrets to the console project as follows, with the neccessary values specified.

{
  "AzureOpenAISettings": {
    "ServiceKey": "<Key>",
    "ServiceName": "<Service-Name>",
    "DeploymentName": "<Model-Deployment-Name>",
    "Endpoint": "https://<Service-Name>.openai.azure.com",
    "MaxCompletionTokens": 500,
    "MaxModelTokens": 8192,
    "Instructions": "Assistant is an intelligent chatbot..."
  },
  "AzureCognitiveSearchSettings": {
    "ServiceKey": "<Key>",
    "ServiceName": "<Service-Name>",
    "Index": "<Index-Name>",
    "SemanticConfig": "<Semantic-Config-Name>",
    "ApiVersion": "<API-Version>",
    "ApiURL": "https://<Service-Name>.search.windows.net/indexes/<Index-Name>/docs?api-version=<API-Version>&search={0}&queryLanguage=en-US&queryType=semantic&captions=extractive&answers=extractive%7Ccount-3&semanticConfiguration=<Semantic-Config-Name>"
  }
}