Utilize prompt engineering in your app

When working with the Azure OpenAI Service, how developers shape their prompt greatly impacts how the generative AI model will respond. Azure OpenAI models are able to tailor and format content, if requested in a clear and concise way. In this exercise, you’ll learn how different prompts for similar content help shape the AI model’s response to better satisfy your requirements.

In scenario for this exercise, you will perform the role of a software developer working on a wildlife marketing campaign. You are exploring how to use generative AI to improve advertising emails and categorize articles that might apply to your team. The prompt engineering techniques used in the exercise can be applied similarly for a variety of use cases.

This exercise will take approximately 30 minutes.

Provision an Azure OpenAI resource

If you don’t already have one, provision an Azure OpenAI resource in your Azure subscription.

Sign into the Azure portal at https://portal.azure.com.
Create an Azure OpenAI resource with the following settings:
- Subscription: Select an Azure subscription that has been approved for access to the Azure OpenAI service
- Resource group: Choose or create a resource group
- Region: Make a random choice from any of the following regions*
  - Australia East
  - Canada East
  - East US
  - East US 2
  - France Central
  - Japan East
  - North Central US
  - Sweden Central
  - Switzerland North
  - UK South
- Name: A unique name of your choice
- Pricing tier: Standard S0
* Azure OpenAI resources are constrained by regional quotas. The listed regions include default quota for the model type(s) used in this exercise. Randomly choosing a region reduces the risk of a single region reaching its quota limit in scenarios where you are sharing a subscription with other users. In the event of a quota limit being reached later in the exercise, there’s a possibility you may need to create another resource in a different region.
Wait for deployment to complete. Then go to the deployed Azure OpenAI resource in the Azure portal.

Deploy a model

Azure provides a web-based portal named Azure AI Foundry portal, that you can use to deploy, manage, and explore models. You’ll start your exploration of Azure OpenAI by using Azure AI Foundry portal to deploy a model.

Note: As you use Azure AI Foundry portal, message boxes suggesting tasks for you to perform may be displayed. You can close these and follow the steps in this exercise.

In the Azure portal, on the Overview page for your Azure OpenAI resource, scroll down to the Get Started section and select the button to go to AI Foundry portal (previously AI Studio).
In Azure AI Foundry portal, in the pane on the left, select the Deployments page and view your existing model deployments. If you don’t already have one, create a new deployment of the gpt-35-turbo-16k model with the following settings:
- Deployment name: A unique name of your choice
- Model: gpt-35-turbo-16k (if the 16k model isn’t available, choose gpt-35-turbo)
- Model version: Use default version
- Deployment type: Standard
- Tokens per minute rate limit: 5K*
- Content filter: Default
- Enable dynamic quota: Disabled
* A rate limit of 5,000 tokens per minute is more than adequate to complete this exercise while leaving capacity for other people using the same subscription.

Explore prompt engineering techniques

Let’s start by exploring some prompt engineering techniques in the Chat playground.

In the Playground section, select the Chat page. The Chat playground page consists of a row of buttons and two main panels (which may be arranged right-to-left horizontally, or top-to-bottom vertically depending on your screen resolution):
- Configuration - used to select your deployment, define system message, and set parameters for interacting with your deployment.
- Chat session - used to submit chat messages and view responses.
Under Deployments, ensure that your gpt-35-turbo-16k model deployment is selected.
Review the default System message, which should be You are an AI assistant that helps people find information.

In the Chat session, submit the following query:

 What kind of article is this?
 ---
 Severe drought likely in California
    
 Millions of California residents are bracing for less water and dry lawns as drought threatens to leave a large swath of the region with a growing water shortage.
    
 In a remarkable indication of drought severity, officials in Southern California have declared a first-of-its-kind action limiting outdoor water use to one day a week for nearly 8 million residents.
    
 Much remains to be determined about how daily life will change as people adjust to a drier normal. But officials are warning the situation is dire and could lead to even more severe limits later in the year.

The response provides a description of the article. However, suppose you want a more specific format for article categorization.

In the Configuration section change the system message to You are a news aggregator that categorizes news articles.

Under the new system message, select the Add section button, and choose Examples. Then add the following example.

User:

 What kind of article is this?
 ---
 New York Baseballers Wins Big Against Chicago
    
 New York Baseballers mounted a big 5-0 shutout against the Chicago Cyclones last night, solidifying their win with a 3 run homerun late in the bottom of the 7th inning.
    
 Pitcher Mario Rogers threw 96 pitches with only two hits for New York, marking his best performance this year.
    
 The Chicago Cyclones' two hits came in the 2nd and the 5th innings but were unable to get the runner home to score.

Assistant:

 Sports

Add another example with the following text.

User:

 Categorize this article:
 ---
 Joyous moments at the Oscars
    
 The Oscars this past week where quite something!
    
 Though a certain scandal might have stolen the show, this year's Academy Awards were full of moments that filled us with joy and even moved us to tears.
 These actors and actresses delivered some truly emotional performances, along with some great laughs, to get us through the winter.
    
 From Robin Kline's history-making win to a full performance by none other than Casey Jensen herself, don't miss tomorrows rerun of all the festivities.

Assistant:

 Entertainment

Use the Apply changes button at the top of the Configuration section to save your changes.

In the Chat session section, resubmit the following prompt:

 What kind of article is this?
 ---
 Severe drought likely in California
    
 Millions of California residents are bracing for less water and dry lawns as drought threatens to leave a large swath of the region with a growing water shortage.
    
 In a remarkable indication of drought severity, officials in Southern California have declared a first-of-its-kind action limiting outdoor water use to one day a week for nearly 8 million residents.
    
 Much remains to be determined about how daily life will change as people adjust to a drier normal. But officials are warning the situation is dire and could lead to even more severe limits later in the year.

The combination of a more specific system message and some examples of expected queries and responses results in a consistent format for the results.

Change the system message back to the default template, which should be You are an AI assistant that helps people find information. with no examples. Then apply the changes.
In the Chat session section, submit the following prompt:
```
# 1. Create a list of animals
# 2. Create a list of whimsical names for those animals
# 3. Combine them randomly into a list of 25 animal and name pairs
```
The model will likely respond with an answer to satisfy the prompt, split into a numbered list. This is an appropriate response, but suppose what you actually wanted was for the model to write a Python program that performs the tasks you described?
Change the system message to You are a coding assistant helping write python code. and apply the changes.

Resubmit the following prompt to the model:

# 1. Create a list of animals
# 2. Create a list of whimsical names for those animals
# 3. Combine them randomly into a list of 25 animal and name pairs

The model should correctly respond with python code doing what the comments requested.

Prepare to develop an app in Visual Studio Code

Now let’s explore the use of prompt engineering in an app that uses the Azure OpenAI service SDK. You’ll develop your app using Visual Studio Code. The code files for your app have been provided in a GitHub repo.

Tip: If you have already cloned the mslearn-openai repo, open it in Visual Studio code. Otherwise, follow these steps to clone it to your development environment.

Start Visual Studio Code.
Open the palette (SHIFT+CTRL+P) and run a Git: Clone command to clone the https://github.com/MicrosoftLearning/mslearn-openai repository to a local folder (it doesn’t matter which folder).
When the repository has been cloned, open the folder in Visual Studio Code.

Note: If Visual Studio Code shows you a pop-up message to prompt you to trust the code you are opening, click on Yes, I trust the authors option in the pop-up.
Wait while additional files are installed to support the C# code projects in the repo.

Note: If you are prompted to add required assets to build and debug, select Not Now.

Configure your application

Applications for both C# and Python have been provided, and both apps feature the same functionality. First, you’ll complete some key parts of the application to enable using your Azure OpenAI resource with asynchronous API calls.

In Visual Studio Code, in the Explorer pane, browse to the Labfiles/03-prompt-engineering folder and expand the CSharp or Python folder depending on your language preference. Each folder contains the language-specific files for an app into which you’re you’re going to integrate Azure OpenAI functionality.
Right-click the CSharp or Python folder containing your code files and open an integrated terminal. Then install the Azure OpenAI SDK package by running the appropriate command for your language preference:

C#:
```
 dotnet add package Azure.AI.OpenAI --version 1.0.0-beta.14
```
Python:
```
 pip install openai==1.13.3
```
In the Explorer pane, in the CSharp or Python folder, open the configuration file for your preferred language
- C#: appsettings.json
- Python: .env
Update the configuration values to include:
- The endpoint and a key from the Azure OpenAI resource you created (available on the Keys and Endpoint page for your Azure OpenAI resource in the Azure portal)
- The deployment name you specified for your model deployment (available in the Deployments page in Azure AI Foundry portal).
Save the configuration file.

Add code to use the Azure OpenAI service

Now you’re ready to use the Azure OpenAI SDK to consume your deployed model.

In the Explorer pane, in the CSharp or Python folder, open the code file for your preferred language, and replace the comment Add Azure OpenAI package with code to add the Azure OpenAI SDK library:

C#: Program.cs
```
 // Add Azure OpenAI package
 using Azure.AI.OpenAI;
```
Python: prompt-engineering.py
```
 # Add Azure OpenAI package
 from openai import AsyncAzureOpenAI
```

In the code file, find the comment Configure the Azure OpenAI client, and add code to configure the Azure OpenAI client:

C#: Program.cs

 // Configure the Azure OpenAI client
 OpenAIClient client = new OpenAIClient(new Uri(oaiEndpoint), new AzureKeyCredential(oaiKey));

Python: prompt-engineering.py

 # Configure the Azure OpenAI client
 client = AsyncAzureOpenAI(
     azure_endpoint = azure_oai_endpoint, 
     api_key=azure_oai_key,  
     api_version="2024-02-15-preview"
     )

In the function that calls the Azure OpenAI model, under the comment Format and send the request to the model, add the code to format and send the request to the model.

C#: Program.cs

 // Format and send the request to the model
 var chatCompletionsOptions = new ChatCompletionsOptions()
 {
     Messages =
     {
         new ChatRequestSystemMessage(systemMessage),
         new ChatRequestUserMessage(userMessage)
     },
     Temperature = 0.7f,
     MaxTokens = 800,
     DeploymentName = oaiDeploymentName
 };
    
 // Get response from Azure OpenAI
 Response<ChatCompletions> response = await client.GetChatCompletionsAsync(chatCompletionsOptions);

Python: prompt-engineering.py

 # Format and send the request to the model
 messages =[
     {"role": "system", "content": system_message},
     {"role": "user", "content": user_message},
 ]
    
 print("\nSending request to Azure OpenAI model...\n")

 # Call the Azure OpenAI model
 response = await client.chat.completions.create(
     model=model,
     messages=messages,
     temperature=0.7,
     max_tokens=800
 )

Save the changes to the code file.

Run your application

Now that your app has been configured, run it to send your request to your model and observe the response. You’ll notice the only difference between the different options is the content of the prompt, all other parameters (such as token count and temperature) remain the same for each request.

In the folder of your preferred language, open system.txt in Visual Studio Code. For each of the interactions, you’ll enter the System message in this file and save it. Each iteration will pause first for you to change the system message.
In the interactive terminal pane, ensure the folder context is the folder for your preferred language. Then enter the following command to run the application.
- C#: dotnet run
- Python: python prompt-engineering.py
Tip: You can use the Maximize panel size (^) icon in the terminal toolbar to see more of the console text.
For the first iteration, enter the following prompts:

System message
```
 You are an AI assistant
```
User message:
```
 Write an intro for a new wildlife Rescue
```
Observe the output. The AI model will likely produce a good generic introduction to a wildlife rescue.
Next, enter the following prompts which specify a format for the response:

System message
```
 You are an AI assistant helping to write emails
```
User message:
```
 Write a promotional email for a new wildlife rescue, including the following: 
 - Rescue name is Contoso 
 - It specializes in elephants 
 - Call for donations to be given at our website
```
Tip: You may find the automatic typing in the VM doesn’t work well with multiline prompts. If that is your issue, copy the entire prompt then paste it into Visual Studio Code.
Observe the output. This time, you’ll likely see the format of an email with the specific animals included, as well as the call for donations.

Next, enter the following prompts that additionally specify the content:

System message

 You are an AI assistant helping to write emails

User message:

 Write a promotional email for a new wildlife rescue, including the following: 
 - Rescue name is Contoso 
 - It specializes in elephants, as well as zebras and giraffes 
 - Call for donations to be given at our website 
 \n Include a list of the current animals we have at our rescue after the signature, in the form of a table. These animals include elephants, zebras, gorillas, lizards, and jackrabbits.

Observe the output, and see how the email has changed based on your clear instructions.

Next, enter the following prompts where we add details about tone to the system message:

System message

 You are an AI assistant that helps write promotional emails to generate interest in a new business. Your tone is light, chit-chat oriented and you always include at least two jokes.

User message:

 Write a promotional email for a new wildlife rescue, including the following: 
 - Rescue name is Contoso 
 - It specializes in elephants, as well as zebras and giraffes 
 - Call for donations to be given at our website 
 \n Include a list of the current animals we have at our rescue after the signature, in the form of a table. These animals include elephants, zebras, gorillas, lizards, and jackrabbits.

Observe the output. This time you’ll likely see the email in a similar format, but with a much more informal tone. You’ll likely even see jokes included!
For the final iteration, we’re deviating from email generation and exploring grounding context. Here you provide a simple system message, and change the app to provide the grounding context as the beginning of the user prompt. The app will then append the user input, and extract information from the grounding context to answer our user prompt.
Open the file grounding.txt and briefly read the grounding context you’ll be inserting.

In your app immediately after the comment Format and send the request to the model and before any existing code, add the following code snippet to read text in from grounding.txt to augment the user prompt with the grounding context.

C#: Program.cs

 // Format and send the request to the model
 Console.WriteLine("\nAdding grounding context from grounding.txt");
 string groundingText = System.IO.File.ReadAllText("grounding.txt");
 userMessage = groundingText + userMessage;

Python: prompt-engineering.py

 # Format and send the request to the model
 print("\nAdding grounding context from grounding.txt")
 grounding_text = open(file="grounding.txt", encoding="utf8").read().strip()
 user_message = grounding_text + user_message

Save the file and rerun your app.

Enter the following prompts (with the system message still being entered and saved in system.txt).

System message

 You're an AI assistant who helps people find information. You'll provide answers from the text provided in the prompt, and respond concisely.

User message:

 What animal is the favorite of children at Contoso?

Tip: If you would like to see the full response from Azure OpenAI, you can set the printFullResponse variable to True, and rerun the app.

Clean up

When you’re done with your Azure OpenAI resource, remember to delete the deployment or the entire resource in the Azure portal at https://portal.azure.com.