Prompt Engineering for LLMs: A Hands-On Azure Tutorial

Key Takeaways

[music] Large language models, often called LLM, are advanced artificial intelligence models that can generate humanlike text. The way someone writes a prompt or instruction can strongly influence the quality of the AI's response. Creating clear and precise prompts is called prompt engineering. This video explores how different approaches to prompt engineering shape the outcome when working with LLMs using the Azure chat playground as the demonstration environment. Before we begin editing prompts, let's walk through how to create a new deployment in the Azure chat playground. First, we navigate to the Azure chat playground by selecting playgrounds on the lefth hand side and then choosing try the chat playground. Next, we'll create a new deployment in the setup by selecting this create new deployment button. Select the model you wish to deploy. In this example, we'll use GPT5 Mini. Select the radio button. Choose confirm. Give it a name. For this example, we'll leave the name as GPT-5- mini, then select deploy. Now, we're ready to craft effective prompts. To begin, consider the importance of prompt clarity. A prompt is a statement or question given to the LLM to start a task. If the prompt is too general, the AI might produce broad or less relevant answers. In this input box, enter the prompt explain climate change. Then select the small arrow to the bottom right of the input box to run the app and see what response we get. As you can see, when we use a general prompt, the LLM gives a long but general answer. There are many facts, but it does not focus on any one aspect. This can make it hard for you to find the specific information you need. More specific prompts can lead to much better results. Adding clear instructions helps the LLM focus. For example, asking the model about a particular effect of climate change narrows its response. Let's try this new prompt. First, we'll clear the output window by selecting the broom in the top right corner, then the blue clear button and enter a new, more detailed prompt. The new prompt instructs the AI to focus on the changes in polar bear habitats over the past decade. This extra detail produces a response that is more targeted and relevant. Providing context and constraints in a prompt, like a time period, location, or specific topic, guides the model to offer more useful and accurate responses. This helps you receive the information you need. Understanding more advanced prompt techniques can further refine AI responses. Two powerful techniques are chain of thought prompting and few shot learning. Before we dive in, let's start fresh with a new prompt. We'll clear the chat history by selecting the broom again in the top right and then choosing clear. Chain of thought prompting is a strategy where the AI is asked to explain its reasoning step by step. This can produce organized and logical answers even for complex questions. Let's enter the prompt, calculate the environmental impact of travel by breaking it into carbon emissions, resource use, and waste. Then compare these factors for air and rail transport. As you can see, we're attempting to have the LLM go through its thought process by calculating the impact first by breaking into carbon emissions, resource use, and waste, and then follow that by comparing these factors. We will select run and see what response we get. After running the prompt, the LLM organizes the answer into sections. We have a section for carbon emissions, a section for resource use, and a section for waste. Then it goes on to make a clear comparison and draw key takeaways in a section down near the bottom. By guiding the model to respond in steps, chain of thought prompting can help ensure the reasoning is transparent and easy to follow. This improves both accuracy and understanding. Let's clear out the page again and talk about fshot learning. Fshot learning is a technique that means giving the LLM several examples before asking it to provide an answer. The model then understands the pattern and tries to follow it in its response. For example, enter the prompt, show two or three examples of product reviews separated by a blank line such as and then we'll provide a format for it to display the reviews in. In this case, the review will have a sentence about the review and then a score with a number divided by a number. Let's select run and see what the model generates. By providing examples, the LLM matched the format and tone displayed. As you can see, we provided these scores for the first two examples and then we asked it to fill in the final score and it did three out of five. Using few shot learning is especially helpful for producing consistent outputs, especially when you want answers in a specific structure or style. Practicing these prompt engineering skills makes it easier to get the most out of large language models. Try different prompt styles during the activity to see how small changes to your prompts affect model responses. Adjust prompts, review the outputs, and compare how the responses improve. For future practice, remember, start with clear and specific prompts, add more detail or structure as needed, use reasoning steps for complex tasks, and supply examples when a certain response style is important. These small adjustments can help LLM deliver responses that match your learning or project goals.