4 Essential LLM Tasks in .NET 10 Using GitHub Models 🚀

Ever wondered how to integrate AI into your .NET applications? In this post, I'll show you 4 fundamental LLM tasks that you can implement in .NET 10 using GitHub Models - and the best part? It's surprisingly simple!

🎯 What We'll Build

1. 💬 Chat Completion - Interactive AI conversations
2. 🏷️ Text Classification - Automatic categorization of user feedback
3. 📝 Text Summarization - Condensing long articles into bullet points
4. 😊 Sentiment Analysis - Understanding customer emotions from reviews

💡 All code is available on GitHub: genai-dotnet-basic_llm_tasks

🔧 Prerequisites

Before we dive in, make sure you have:

• ✅ .NET 10 SDK installed
• ✅ Visual Studio 2022 or VS Code
• ✅ GitHub Personal Access Token with access to GitHub Models

Getting Started with GitHub Models

GitHub Models provides free access to powerful AI models! Here's how to get started:

1. Get a GitHub PAT (Personal Access Token)

   • Go to GitHub Settings → Developer Settings → Tokens
   • Generate a token with repo access
   • Save it securely!

2. Configure User Secrets (Keep your token safe!)
   

cd YourProject
dotnet user-secrets set "GitHubModels:Token" "your-github-token-here"

📚 Learn more: Check out the official GitHub Models documentation for detailed setup instructions.

1️⃣ Chat Completion - Your AI Assistant 💬

Let's start with the basics: having a conversation with AI. This project demonstrates both non-streaming (get complete response) and streaming (real-time, token-by-token) responses.

Architecture at a Glance

Configuration → Authentication → AI Client → Chat Response

The Code

using Microsoft.Extensions.AI;
using Microsoft.Extensions.Configuration;
using OpenAI;
using System.ClientModel;

// Load configuration
IConfigurationRoot configuration = new ConfigurationBuilder()
    .SetBasePath(Directory.GetCurrentDirectory())
    .AddJsonFile("appsettings.json", optional: false)
    .AddUserSecrets<Program>()
    .Build();

// Authenticate with GitHub Models
ApiKeyCredential credential = new ApiKeyCredential(
    configuration["GitHubModels:Token"] 
    ?? throw new InvalidOperationException("Token not found!"));

// Initialize AI client
IChatClient chatClient = new OpenAIClient(credential, new OpenAIClientOptions
{
    Endpoint = new Uri("https://models.inference.ai.azure.com")
})
.GetChatClient("gpt-4o-mini")
.AsIChatClient();

// Get streaming response
var responseStream = chatClient.GetStreamingResponseAsync(
    "Explain quantum computing in simple terms", 
    new ChatOptions { Temperature = 0.7f, MaxOutputTokens = 300 }
);

Console.Write("AI: ");
await foreach (var message in responseStream)
{
    Console.Write(message.Text);
}

🎨 What Makes It Cool?

• Real-time streaming for better UX
• Token usage tracking for cost management
• Configurable temperature (0.0 = focused, 1.0 = creative)

Use cases: Chatbots, Q&A systems, code assistants

2️⃣ Text Classification - Smart Categorization 🏷️

Automatically categorize user feedback into Complaint, Suggestion, Praise, or Other. Perfect for customer support automation!

The Magic Prompt

var classificationPrompt = @"Classify the text into one category: 
Complaint, Suggestion, Praise, or Other.
Respond with only the category name.
Text: '{0}'";

var userFeedback = "The app keeps crashing when I upload photos!";

var response = await chatClient.GetResponseAsync(
    string.Format(classificationPrompt, userFeedback),
    new ChatOptions 
    { 
        Temperature = 0.1f,  // Low temp for consistency
        MaxOutputTokens = 50 
    }
);

Console.WriteLine($"Classification: {response.Text}");
// Output: Complaint

💡 Pro Tips

• Use low temperature (0.1) for consistent classifications
• Keep MaxOutputTokens minimal since you only need one word
• Perfect for routing support tickets!

Real-world impact: Route 1000s of customer messages automatically 📮

3️⃣ Text Summarization - TL;DR Generator 📝

Turn lengthy articles into concise bullet points. Save reading time and improve information accessibility!

Quick Example

var summaryPrompt = @"Summarize the following text as bullet points.
Provide 2-5 key points.
Text: '{0}'";

var longText = @"The quarterly earnings report shows our company 
exceeded expectations with a 15% revenue increase. Growth was driven 
by Asia-Pacific sales and new product launches. Profit margin 
improved from 12% to 14%, and the board approved a 5% dividend increase.";

var summary = await chatClient.GetResponseAsync(
    string.Format(summaryPrompt, longText),
    new ChatOptions { Temperature = 0.1f, MaxOutputTokens = 150 }
);

Console.WriteLine($"Summary:\n{summary.Text}");

Output

• Company revenue increased 15% year-over-year, beating expectations
• Growth driven by Asia-Pacific sales and new product launches  
• Profit margin improved from 12% to 14%
• Board approved 5% dividend increase for shareholders

Use cases: News aggregation, document processing, meeting notes

4️⃣ Sentiment Analysis - Understand Customer Emotions 😊

Go beyond simple positive/negative! This returns structured JSON with sentiment breakdown, positive/negative aspects, and emotional tone.

The Smart Prompt

var sentimentPrompt = @"Analyze the sentiment of this product review.
Return as JSON with this structure:
{
  ""overallSentiment"": ""Positive/Negative/Neutral/Mixed"",
  ""positiveAspects"": [""list of good things""],
  ""negativeAspects"": [""list of bad things""],
  ""emotionalTone"": ""description of tone""
}
Review: '{0}'";

var review = @"This laptop is amazing! Battery lasts all day, 
screen is crystal clear. Best purchase this year!";

var analysis = await chatClient.GetResponseAsync(
    string.Format(sentimentPrompt, review),
    new ChatOptions { Temperature = 0.1f, MaxOutputTokens = 200 }
);

Console.WriteLine($"Analysis:\n{analysis.Text}");

Sample Output

{
  "overallSentiment": "Positive",
  "positiveAspects": [
    "Excellent battery life lasting all day",
    "Crystal clear screen quality",
    "Overall satisfaction with purchase"
  ],
  "negativeAspects": [],
  "emotionalTone": "Enthusiastic and highly satisfied"
}

🎯 Why JSON?

• Easy to parse programmatically
• Structured data for dashboards
• Multi-dimensional insights in one request

Business value: Build customer sentiment dashboards, track product issues, identify improvement areas 📊

🚀 Getting Started

Ready to try it yourself?

# Clone the repo
git clone https://github.com/Rahul1994jh/genai-dotnet-basic_llm_tasks
cd genai-dotnet-basic_llm_tasks

# Setup your GitHub token
dotnet user-secrets set "GitHubModels:Token" "your-token-here"

# Run any project
dotnet run --project TextCompletion
dotnet run --project Classification
dotnet run --project Summarization
dotnet run --project SentimentAnalysis

Each project has a detailed README with:

• Architecture diagrams
• Customization guides
• Troubleshooting tips
• Best practices

🎓 Key Takeaways

💰 Cost Optimization Tips

• Use gpt-4o-mini for general tasks (10x cheaper!)
• Set MaxOutputTokens appropriately
• Use low temperature for predictable tasks
• Monitor token usage in production

🔐 Security Best Practices

✅ DO:

• Store tokens in User Secrets (development)
• Use Azure Key Vault (production)
• Use .gitignore for secrets
• Rotate tokens regularly

❌ DON'T:

• Hardcode API keys in code
• Commit secrets to Git
• Share tokens publicly

🎯 What's Next?

Now that you've mastered the basics, here are some ideas to level up:

1. Combine tasks: Classify → Analyze Sentiment → Summarize
2. Add conversation history for multi-turn chats
3. Build a web API with ASP.NET Core
4. Create a Blazor UI for interactive demos
5. Add function calling for tool integration

📚 Resources

• 📦 GitHub Repo: genai-dotnet-basic_llm_tasks
• 🤖 GitHub Models: Official Documentation
• 🔧 Microsoft.Extensions.AI: NuGet Package
• 📖 .NET 10: What's New

💬 Let's Connect!

Found this helpful? Have questions or ideas?

• ⭐ Star the repo on GitHub
• 💬 Drop a comment below
• 🐦 Share with your network

Happy coding! 🎉

Built with ❤️ using .NET 10, Microsoft.Extensions.AI, and GitHub Models

dotnet #ai #machinelearning #csharp #github #llm #artificialintelligence #programming