Exercise - Create native functions

10 minutes

In this exercise, you build a travel agent that helps users find and book flights. Let's get started!

Prepare your development environment

A starter project is available for you to use for this exercise. Use the following steps to set up the starter project:

Important

You must have Visual Studio Code and the .NET Framework 8.0 installed to complete these steps. You may also need to install the Visual Studio Code C# Dev Kit extension.

Open Visual Studio Code.
Under the Visual Studio Code Start section, select Clone Git Repository.
In the URL bar, enter https://github.com/MicrosoftLearning/MSLearn-Develop-AI-Agents-with-Azure-OpenAI-and-Semantic-Kernel-SDK.git
In the File Explorer, create a new folder in a location that is easy to find and remember, such as a folder in your Desktop.
Click the Select as Repository Destination button.

You need to be signed in to GitHub to successfully clone the project.
Open the project in Visual Studio Code.
Open the appsettings.json file and update the values with your Azure OpenAI Services modelId, endpoint, and API key.
```
{
    "modelId": "gpt-35-turbo-16k",
    "endpoint": "",
    "apiKey": ""
}
```
In Explorer, right-click the M03-create-semantic-kernel-plugins/M03-Project folder and click Open in Integrated Terminal.
Expand the M03-create-semantic-kernel-plugins/M03-Project folder.

You should see a "Program.cs" file.

Now you're ready to begin the exercise. Good luck!

Create a flight booking plugin

In this task, you create a plugin class for your travel agent. The class includes functions to search for available flights and book a selected flight for the user. A flights.json file is included in the project which contains some sample flight information.

Open the FlightBookingPlugin.cs in the M03-Project folder.

Create a function that searches for flights by entering the following code:

public List<FlightModel> SearchFlights(string destination, string departureDate)
{
    // Filter flights based on destination
    return flights.Where(flight =>
        flight.Destination.Equals(destination, StringComparison.OrdinalIgnoreCase) &&
        flight.DepartureDate.Equals(departureDate)).ToList();
}

Add attributes to your function that will help the AI understand how to call the function:

[KernelFunction("search_flights")]
[Description("Searches for available flights based on the destination and departure date in the format YYYY-MM-DD")]
[return: Description("A list of available flights")]
public List<FlightModel> SearchFlights(string destination, string departureDate)

Add a function that updates the flight as booked for the user:

public string BookFlight(int flightId)
{
    var flight = flights.FirstOrDefault(f => f.Id == flightId);
    if (flight == null)
    {
        return "Flight not found. Please provide a valid flight ID.";
    }

    if (flight.IsBooked)
    {
        return $"You've already booked this flight.";
    }

    flight.IsBooked = true;
    SaveFlightsToFile();

    return $"Flight booked successfully! Airline: {flight.Airline}, Destination: {flight.Destination}, Departure: {flight.DepartureDate}, Price: ${flight.Price}.";
}

Add attributes to your function that will help the AI understand how to call the function:

[KernelFunction("book_flight")]
[Description("Books a flight based on the flight ID provided")]
[return: Description("Booking confirmation message")]

Navigate to the Program.cs file.

Add your plugin to the kernel with the following code:

kernel.Plugins.AddFromType<FlightBookingPlugin>("FlightBooking");

Add the following code to enable automatic function calling:

OpenAIPromptExecutionSettings openAIPromptExecutionSettings = new() 
{
    FunctionChoiceBehavior = FunctionChoiceBehavior.Auto()
};

Create a chat history object and add a prompt to guide the LLM:

var history = new ChatHistory();
history.AddSystemMessage("The year is 2025 and the current month is January");

Enter the following code to a prompt to trigger the SearchFlights function:

AddUserMessage("Find me a flight to Tokyo on the 19");
await GetReply();
GetInput();
await GetReply();

To test your code, open the M03-Project folder in the terminal and enter dotnet run

You should see output similar to the following:

User: Find me a flight to Tokyo on the 19
Assistant: I found a flight to Tokyo on January 19th. 

- Airline: Air Japan
- Price: $1200

Would you like to book this flight?

Enter "Yes" in the terminal to invoke the BookFlight function

You should see output similar to the following:

User: Yes
Assistant: Congratulations! Your flight to Tokyo on January 19th with Air Japan has been successfully booked. The total price for the flight is $1200.

Now you successfully added plugin functions to your kernel that can be invoked automatically. Great work!

Exercise - Create native functions

Prepare your development environment

Create a flight booking plugin

Feedback