Prompt Engineering

Summary

Prompt Engineering is the practice of designing, testing, and optimizing prompts (text written by a human to talk to an AI) for language models (LLMs), with the goal of obtaining more accurate, useful, and controlled responses.

Why does it matter?

To get the answer you want from an AI, which looks for patterns and structures in information, the human has to provide a good explanation. If what you want is not explicit, the AI starts to hallucinate: it returns an answer that is often incorrect, mixing up topics and stating falsehoods that can harm the process you were counting on.

A well-crafted prompt can turn a generic answer into a highly specialized solution.

Prompt-building process

1. Understand the model: Know how the chosen model (Gemini, ChatGPT, Copilot) works, including its limitations and capabilities.
2. Write the prompt:
   • Clarity: Avoid ambiguity in your prompt.
   • Context: Provide as much relevant information as possible.
   • Format: Use lists, examples, instructions, etc.
3. Iterate and adjust:
   • Test variations and compare the results.
   • Measure the quality of the answers.
   • Refine based on feedback or metrics.

Prompting techniques

• Few-shot prompting: Give examples in the prompt.
• Chain-of-thought prompting: Lead the model to think step by step.
• Zero-shot prompting: Ask direct questions without any example.
• Prompt templates: Create reusable structures for recurring tasks.

Zero-shot

• No example is provided.
• Suited to simpler answers, nothing specific.
• Direct questions without preparation.

Write a caption for a photo of an artisan coffee shop.

One-shot

• One example is provided.
• Gives the AI an example of something specific.
• The AI follows the given model to create the text.

Write a caption for a photo of an artisan coffee shop.
Example caption: Nothing beats starting the day with a special coffee made with care.
Now write a similar caption for a new post.

Few-shot

• A few examples are provided.
• Prepare 3 or more examples so the model can build a pattern that guides its answer.

Explain what discipline is following these models:

Example 1:
Question: What is Productivity?
Answer: The ability to produce more results with less effort.

Example 2:
Question: What is Focus?
Answer: The skill of directing your attention to a single task.

Example 3:
Question: What is Motivation?
Answer: Motivation is the inner energy that drives a person to act.

Role Play technique

• Assigns a role to the AI.
• It filters its knowledge to focus on that specific role.
• The type of answer changes to match the role.
• Combining a role with examples and more details delivers a complete answer with fewer mistakes.

Act as a Software Engineer with knowledge of C#, .NET, and ASP.NET Core.
Explain directly how Garbage Collection works in .NET and show me an example of using it.

CARE Framework

CARE is a structured way to build more effective prompts. Here are the steps:

Context → Set the scene, explain the background of the problem, and give the AI any details you find useful for thinking about it.

Ask → Be clear about what you want and what kind of answer you expect from the AI.

Rules → Define the desired limits: tone, length, style, and the expected audience for the answer.

Example → If you know of an example that can help the AI reach better results, write it down; the more examples, the better the outcome.

Example

This is a very vague prompt:

Help me write an update message for my team

This is a prompt using the CARE pattern:

You are helping a project lead communicate an update to remote teams. Write a friendly and concise message of at most 100 words summarizing the project timeline. Use a tone similar to “Quick update for everyone - here is what has changed…”.