📚 AI Foundation

📖 Simple Chat

using Azure;
using Azure.AI.OpenAI;
using OpenAI;
using OpenAI.Chat;

var endpoint = new Uri("https://xxxx.azure.com/");
var deploymentName = "gpt-35-turbo";
var apiKey = "xxx";

OpenAIClient client = new OpenAIClient(apiKey); // OpenAI


AzureOpenAIClient azureClient = new(
    endpoint,
    new AzureKeyCredential(apiKey)); // Azure OpenAI

ChatClient chatClient = azureClient.GetChatClient(deploymentName);

var requestOptions = new ChatCompletionOptions()
{
    MaxOutputTokenCount = 4096,
    Temperature = 1.0f,
    TopP = 1.0f,
};

List<ChatMessage> messages = new List<ChatMessage>()
{
    new SystemChatMessage("You are a helpful assistant. Be brief!"),
    new UserChatMessage("I am going to Paris, what should I see?"),
};

var response = chatClient.CompleteChat(messages, requestOptions);

switch(response.Value.FinishReason)
{
    case ChatFinishReason.Stop: // Response is ready
        Console.WriteLine(response.Value.Content[0].Text);
        break;

    case ChatFinishReason.Length: // hit max_tokens or context window
        Console.WriteLine("[truncated] " + response.Value.Content[0].Text);
        // ➜ Mitigate: raise MaxOutputTokenCount, shorten history, summarize older turns.
        break;

    case ChatFinishReason.ContentFilter:
        Console.WriteLine("[blocked] Output filtered by Azure content filters.");
        break;

    default:
        Console.WriteLine("[unknown finish] " + response.Value.Content[0].Text);
        break;
}

Console.ReadKey();

📖 Streaming

When we compare our chat application with the official ChatGPT interface, there is a big difference. Even though the generation speed was pretty much the same, our app still felt a lot slower. Why?
Because our app waited until the entire poem was ready before showing it. And ChatGPT showed the text as it was being written word by word. This is called streaming. ChatGPT displays the response in chunks, allowing us to start reading the answer much earlier.
The current generation speed of GPT-4 is slightly faster than what a human can read, but the GPT-3.5 is faster than GPT-4. This means that if we stream the response, the user can essentially follow along and doesn't perceive the AI as slow.

As cool as streaming is, it has its disadvantages:

• It's more complex to implement. Instead of simply receiving and displaying a complete response

• We have to handle partial responses, keep track of the chunks, handle intermediate errors, and so on

• Requires more bandwidth as every chunk comes with quite a lot of overhead and only one or two useful words worth of content.

Should we use streaming in your project? The answer, as it's pretty much always is, it depends.

• The model

• The load on the server

• he amount of text you need to generate

• User's expectations or mindset.

Program.cs

using System;
using System.Collections.Generic;
using System.Text;
using System.Threading.Tasks;
using Azure;
using Azure.AI.OpenAI;

class Program
{
    static async Task Main()
    {
        var endpoint = new Uri("https://xxxx.azure.com/");
        var apiKey = "AZURE_API_KEY";
        var deploymentName = "gpt-35-turbo";

        var azureClient = new AzureOpenAIClient(endpoint, new AzureKeyCredential(apiKey));
        ChatClient chatClient = azureClient.GetChatClient(deploymentName);

        var messages = new List<ChatMessage>
        {
            new SystemChatMessage("You are a helpful assistant. Be brief!"),
            new UserChatMessage("I am going to Paris, what should I see?")
        };

        Console.OutputEncoding = Encoding.UTF8;
        var sb = new StringBuilder();

        Console.Write("Assistant: ");

        // ✅ Stream results asynchronously
        await foreach (StreamingChatCompletionUpdate update in chatClient.CompleteChatStreamingAsync(messages))
        {
            foreach (var part in update.ContentUpdate)
            {
                Console.Write(part.Text);
                sb.Append(part.Text);
            }
        }

        Console.WriteLine("\n\nFull message received:");
        Console.WriteLine(sb.ToString());
    }
}

Full Chat Loop

using System;
using System.Collections.Generic;
using System.Text;
using System.Threading.Tasks;
using Azure;
using Azure.AI.OpenAI;

class Program
{
    static async Task Main()
    {
        var endpoint = new Uri("https://mkou-mhde9425-eastus2.cognitiveservices.azure.com/");
        var apiKey = "AZURE_API_KEY";
        var deploymentName = "gpt-35-turbo";

        var azureClient = new AzureOpenAIClient(endpoint, new AzureKeyCredential(apiKey));
        var chatClient = azureClient.GetChatClient(deploymentName);

        // 🧠 Keeps the whole conversation (system + user + assistant)
        var messages = new List<ChatMessage>
        {
            new SystemChatMessage("You are a helpful and concise assistant.")
        };

        Console.OutputEncoding = Encoding.UTF8;
        Console.WriteLine("=== Chat started (type 'exit' to quit) ===");

        while (true)
        {
            Console.Write("\nYou: ");
            string? userInput = Console.ReadLine();

            if (string.IsNullOrWhiteSpace(userInput))
                continue;
            if (userInput.Equals("exit", StringComparison.OrdinalIgnoreCase))
                break;

            // Add user message to conversation history
            messages.Add(new UserChatMessage(userInput));

            Console.ForegroundColor = ConsoleColor.Yellow;

            Console.Write("Assistant: ");
            var sb = new StringBuilder();

            // 🔁 Stream assistant reply
            await foreach (StreamingChatCompletionUpdate update in chatClient.CompleteChatStreamingAsync(messages))
            {
                foreach (var part in update.ContentUpdate)
                {
                    Console.Write(part.Text);
                    sb.Append(part.Text);
                }
            }

            Console.ForegroundColor = ConsoleColor.White;

            Console.WriteLine();

            // Add assistant message to conversation history
            messages.Add(new AssistantChatMessage(sb.ToString()));
        }

        Console.WriteLine("\nChat ended.");
    }
}