Prompt Engineering Challenges

Introduction

Targeted challenges to sharpen your prompting skills. Each one mirrors a real production problem and forces you to think about evaluation, edge cases and cost.

Understanding the topic

Core concepts to understand:

• Extract structured JSON from messy emails.
• Build a one-shot summariser at three lengths (50, 150, 500 words).
• Convert legal text to plain English without losing meaning.
• Generate code in three languages from one spec.
• Build a 'translate to user's tone' rewriter.
• Solve a logic puzzle with chain-of-thought, then without.
• Create a fact-checked answer with confidence score.
• Design a prompt that refuses gracefully on out-of-scope questions.

Syntax reference

Visual workflow / architecture:

Challenge → Test set (10 inputs) → Prompt v1 → Score
                                       │
                                       ▼
                                 Iterate → Score
                                       │
                                       ▼
                                 Best prompt

Real-world use

These are real interview tasks at Anthropic, OpenAI, Cohere, Cursor.

Best practices

• Always create a small test set first.
• Track accuracy + cost across iterations.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. Walk through how you'd prompt-engineer a JSON extractor.
• Q2. How would you evaluate it?
• Q3. How do you handle ambiguous inputs?