Prompt Engineering Fundamentals

# CHAPTER 6

Prompt Engineering Fundamentals #

1. Introduction #

2. Learning Objectives #

• Define Prompt Engineering and why it is a critical skill.

• Structure a complex prompt using Personas, Context, and Formatting.

• Distinguish between Zero-shot and Few-shot prompting.

• Apply the "Chain of Thought" technique to improve AI reasoning.

3. Beginner-Friendly Explanation #

4. The Anatomy of a Perfect Prompt #

1. 1. Persona/Role: "Act as a Senior Python Developer."

1. 2. Task/Instruction: "Review this code and find the security bugs."

1. 3. Context: "This code is for a banking application where user privacy is critical."

1. 4. Format Constraints: "Output your findings in a Markdown table, ordered by severity."

5. Zero-Shot vs Few-Shot Prompting #

• Zero-Shot Prompting: Asking the AI to do something without giving it any examples. *Example:* "Classify the sentiment of this text: 'I hated the food.' -> "

• Few-Shot Prompting: Giving the AI a few examples of the exact input and expected output *before* asking it to perform the task. This drastically increases accuracy because it mathematically anchors the AI to your specific format.

6. Chain of Thought Prompting (CoT) #

7. Python Example: Implementing Few-Shot #

from openai import OpenAI
client = OpenAI()

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": "You are a customer service bot."},
        # Few-Shot Examples to teach the bot the format
        {"role": "user", "content": "Where is my order?"},
        {"role": "assistant", "content": "ACTION: CHECK_TRACKING | MESSAGE: I will look that up for you!"},
        {"role": "user", "content": "I want a refund."},
        {"role": "assistant", "content": "ACTION: INITIATE_RETURN | MESSAGE: I'm sorry to hear that. Let's start the return."},
        # The actual live user prompt
        {"role": "user", "content": "My shirt arrived torn!"}
    ]
)

print(response.choices[0].message.content)
# Expected Output: ACTION: INITIATE_RETURN | MESSAGE: I apologize your shirt was damaged. Let's fix that.

8. Mini Project #

9. Best Practices #

• Delimiters: Use specific symbols (like triple quotes """ or XML tags <data>) to separate your instructions from the text you want the AI to analyze. Example: *Summarize the text enclosed in triple quotes: """[Paste Text]"""*

10. Common Mistakes #

• Being Vague: Do not assume the AI knows what you are thinking. If you want the output to be exactly 3 paragraphs, say "Write exactly 3 paragraphs." If you want it to avoid using complex jargon, explicitly state "Do not use technical jargon."

11. Exercises #

1. 1. Explain how adding the phrase "Think step-by-step" (Chain of Thought) helps an LLM solve a complex math word problem.

12. MCQs with Answers #

13. Interview Questions #

• Q: Break down the difference between Zero-shot, One-shot, and Few-shot prompting.

• Q: As a Prompt Engineer, how would you construct a prompt to ensure an LLM extracts data from a messy text file and strictly formats it into a valid JSON object without adding conversational filler text?

14. FAQs #

15. Summary #

16. Next Chapter Recommendation #