Summarize text using Azure AI chat app with .NET

Get started with Semantic Kernel by creating a simple .NET 8 console chat application. The application will run locally and use the OpenAI gpt-35-turbo model deployed into an Azure OpenAI account. Follow these steps to provision Azure OpenAI and learn how to use Semantic Kernel.

Get started with the .NET Azure OpenAI SDK by creating a simple .NET 8 console chat application. The application will run locally and use the OpenAI gpt-35-turbo model deployed into an Azure OpenAI account. Follow these steps to provision Azure OpenAI and learn how to use the .NET Azure OpenAI SDK.

1. From a terminal or command prompt, navigate to the semantic-kernel\01-HikeBenefitsSummary directory.

1. From a terminal or command prompt, navigate to the azure-openai-sdk\01-HikeBenefitsSummary directory.

Understanding the code

Our application uses the Microsoft.SemanticKernel package, which is available on NuGet, to send and receive requests to an Azure OpenAI service deployed in Azure.

The entire application is contained within the Program.cs file. The first several lines of code loads up secrets and configuration values that were set in the dotnet user-secrets for you during the application provisioning.

// == Retrieve the local secrets saved during the Azure deployment ==========
var config = new ConfigurationBuilder()
    .AddUserSecrets<Program>()
    .Build();
string endpoint = config["AZURE_OPENAI_ENDPOINT"];
string deployment = config["AZURE_OPENAI_GPT_NAME"];
string key = config["AZURE_OPENAI_KEY"];

// Create a Kernel containing the Azure OpenAI Chat Completion Service
Kernel kernel = Kernel.CreateBuilder()
    .AddAzureOpenAIChatCompletion(deployment, endpoint, key)
    .Build();

The Kernel class facilitates the requests and responses with the help of AddAzureOpenAIChatCompletion service.

Kernel kernel = Kernel.CreateBuilder()
    .AddAzureOpenAIChatCompletion(deployment, endpoint, key)
    .Build();

Once the Kernel is created, we read the contents of the file benefits.md and create a prompt to ask the the model to summarize that text.

// Create and print out the prompt
string prompt = $"""
    Please summarize the the following text in 20 words or less:
    {File.ReadAllText("benefits.md")}
    """;
Console.WriteLine($"user >>> {prompt}");

To have the model generate a response based off prompt, use the InvokePromptAsync function.

// Submit the prompt and print out the response
string response = await kernel.InvokePromptAsync<string>(prompt, new(new OpenAIPromptExecutionSettings() { MaxTokens = 400 }));
Console.WriteLine($"assistant >>> {response}");

Customize the text content of the file or the length of the summary to see the differences in the responses.

Understanding the code

Our application uses the Azure.AI.OpenAI client SDK, which is available on NuGet, to send and receive requests to an Azure OpenAI service deployed in Azure.

The entire application is contained within the Program.cs file. The first several lines of code loads up secrets and configuration values that were set in the dotnet user-secrets for you during the application provisioning.

// == Retrieve the local secrets saved during the Azure deployment ==========
var config = new ConfigurationBuilder()
    .AddUserSecrets<Program>()
    .Build();

string openAIEndpoint = config["AZURE_OPENAI_ENDPOINT"];
string openAIDeploymentName = config["AZURE_OPENAI_GPT_NAME"];
string openAiKey = config["AZURE_OPENAI_KEY"];

// == Creating the AIClient ==========
var endpoint = new Uri(openAIEndpoint);
var credentials = new AzureKeyCredential(openAiKey);

The OpenAIClient class facilitates the requests and responses. ChatCompletionOptions specifies parameters of how the model will respond.

var openAIClient = new OpenAIClient(endpoint, credentials);

var completionOptions = new ChatCompletionsOptions
{
    MaxTokens = 400,
    Temperature = 1f,
    FrequencyPenalty = 0.0f,
    PresencePenalty = 0.0f,
    NucleusSamplingFactor = 0.95f, // Top P
    DeploymentName = openAIDeploymentName
};

Once the OpenAIClient client is created, we read the content of the file benefits.md. Then using the ChatRequestUserMessage class we can add to the model the request to summarize that text.

string userRequest = """
Please summarize the the following text in 20 words or less:
""" + markdown;

completionOptions.Messages.Add(new ChatRequestUserMessage(userRequest));
Console.WriteLine($"\n\nUser >>> {userRequest}");

To have the model generate a response based off the user request, use the GetChatCompletionsAsync function.

ChatCompletions response = await openAIClient.GetChatCompletionsAsync(completionOptions);
ChatResponseMessage assistantResponse = response.Choices[0].Message;
Console.WriteLine($"\n\nAssistant >>> {assistantResponse.Content}");

Customize the text content of the file or the length of the summary to see the differences in the responses.