AI Integration for .NET Developers: Azure and GitHub Tools in 2024

TL;DR
Azure AI and GitHub Copilot are rapidly evolving, offering .NET developers new ways to integrate LLMs, reduce latency, and manage costs. Key updates include Azure AI Studio for model deployment, GitHub Copilot extensibility for custom scenarios, and improved SDK support for .NET apps. This post covers practical steps and considerations.

Azure AI Studio and .NET SDK Integration

Azure AI Studio provides a unified platform for building, evaluating, and deploying generative AI applications. For .NET developers, the Azure.AI.OpenAI SDK is the primary tool for integrating these capabilities. Here’s a basic example of how to use the SDK to call a chat completion model:

using Azure;
using Azure.AI.OpenAI;

OpenAIClient client = new OpenAIClient(
    new Uri("https://your-resource.openai.azure.com/"),
    new AzureKeyCredential("your-api-key"));

Response<ChatCompletions> response = await client.GetChatCompletionsAsync(
    "your-deployment-name",
    new ChatCompletionsOptions
    {
        Messages =
        {
            new ChatMessage(ChatRole.System, "You are a helpful assistant."),
            new ChatMessage(ChatRole.User, "How do I use Azure OpenAI with .NET?")
        }
    });

Console.WriteLine(response.Value.Choices[0].Message.Content);

Considerations:

  • Cost: Be mindful of token usage, especially with high-volume applications. Use streaming where possible to improve perceived latency.
  • Latency: Choose a region close to your users. For critical paths, implement caching strategies.

GitHub Copilot for .NET Development

GitHub Copilot continues to enhance productivity for .NET developers. Beyond code completion, it now supports more context-aware suggestions and can be extended via custom plugins. To get the most out of Copilot:

  • Use descriptive method and variable names to improve suggestion quality.
  • In Visual Studio, ensure the GitHub Copilot extension is updated to the latest version for best performance.

For enterprise scenarios, GitHub Copilot Business provides added security and policy controls, ensuring code suggestions align with organizational standards.

Azure OpenAI and Semantic Kernel

Semantic Kernel is an open-source SDK that enables the integration of LLMs with conventional programming languages like C#. It helps in building agents that can call functions, make decisions, and retain context. Example of planning a sequence with Semantic Kernel:

using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Planning;

var kernel = Kernel.CreateBuilder()
    .AddAzureOpenAIChatCompletion("your-deployment-name", "endpoint", "api-key")
    .Build();

// Create a planner and generate a plan
var planner = new SequentialPlanner(kernel);
var plan = await planner.CreatePlanAsync("Write a poem about .NET and AI");

// Execute the plan
var result = await plan.InvokeAsync(kernel);
Console.WriteLine(result);

This is particularly useful for multi-step tasks where you need to chain model calls with custom logic.

Cost and Performance Optimization

When working with Azure OpenAI, keep these in mind:

  • Use the max_tokens parameter to limit response length and control cost.
  • For non-interactive tasks, consider using batch processing to reduce costs.
  • Evaluate using smaller, cheaper models (e.g., GPT-3.5-turbo) for simpler tasks where feasible.

Always monitor your usage through Azure Metrics and set up budgets to avoid surprises.

Getting Started Checklist

  1. Set up an Azure OpenAI resource in your desired region.
  2. Deploy a model (e.g., gpt-35-turbo or gpt-4) noting the deployment name.
  3. Install necessary NuGet packages: Azure.AI.OpenAI and Microsoft.SemanticKernel.
  4. Experiment in Azure AI Studio for quick prototyping and evaluation.
  5. Integrate with your .NET app using the SDK, starting with non-critical functions.
  6. Implement telemetry to track token usage, latency, and errors.

Further reading

  • Azure AI Studio documentation: https://learn.microsoft.com/azure/ai-studio/
  • Azure.AI.OpenAI SDK for .NET: https://learn.microsoft.com/dotnet/api/azure.ai.openapi
  • Semantic Kernel for .NET: https://learn.microsoft.com/semantic-kernel/overview/
  • GitHub Copilot for Business: https://docs.github.com/en/copilot/overview-of-github-copilot/about-github-copilot-for-business