The Prompt Engineering Roadmap
A structured 6-week journey from complete beginner to competent, creative prompt engineer. Click each phase to explore.
Core Concept: Understand how Large Language Models (LLMs) work without complex math.
Key Topics:
- Probabilistic text prediction — think of it as autocomplete on steroids.
- The importance of context and instructions — the AI has no memory between conversations unless you provide it.
- Recognizing hallucinations and limitations — the AI can sound confident while being wrong.
Core Concept: A structured approach to writing prompts so you stop guessing and start engineering.
The CREATE framework gives you six elements to include in every prompt:
- Context — Who is the AI, and what is the background situation?
- Role — What persona should the AI adopt?
- Explicit Instructions — What exact steps or tasks need to be performed?
- Audience — Who is the final output for?
- Tone — Professional, casual, academic, or persuasive?
- Examples / Output Format — What should the response look like?
Navigate to the "CREATE Framework" section for an interactive deep-dive.
Core Concept: Move beyond basic Q&A to prompt optimization strategies.
- Few-Shot Prompting — Provide a few examples of desired inputs and outputs to set a pattern.
- Chain-of-Thought (CoT) Prompting — Instruct the AI to "think step-by-step" before answering to improve logic and accuracy.
- Constraint Setting — Explicitly tell the AI what not to do (e.g., "Do not use jargon," "Keep it under 200 words").
Core Concept: Apply prompt engineering to real-world workflows.
Practical Applications:
- Summarizing research papers
- Drafting professional emails
- Brainstorming and ideation
- Writing and debugging code
- Creating educational content
The CREATE Framework
Click each letter to learn about that element. Together, they form a complete, professional prompt.
Advanced Techniques for Beginners
Three powerful strategies that elevate your prompts from basic to professional.
Instead of just describing what you want, you provide the AI with example input-output pairs so it can recognize the pattern and replicate it.
Classify the sentiment of each review as Positive, Negative, or Neutral.
Review: "The food was amazing and the service was fast."
Sentiment: Positive
Review: "Terrible experience, waited 45 minutes for cold food."
Sentiment: Negative
Review: "It was okay, nothing special but not bad either."
Sentiment: Neutral
Review: "Absolutely loved the new update, works perfectly now!"
Sentiment:Why it works: The examples teach the AI the exact classification pattern, format, and granularity you expect — far more reliable than just saying "classify the sentiment."
By asking the AI to reason step-by-step before giving a final answer, you dramatically reduce errors in logic, math, and multi-step reasoning.
If a store sells apples for $2 each, but offers a 10% discount on purchases over $10, and John buys 6 apples, how much does he pay?Calculate the total cost of 6 apples priced at $2 each, keeping in mind there is a 10% discount for purchases over $10. Think step-by-step before providing the final answer.Explicitly telling the AI what not to do is often more powerful than telling it what to do. Constraints prevent unwanted behaviors.
- "Do not use technical jargon. Explain as if to a 12-year-old."
- "Keep the response under 200 words."
- "Do not include an introduction or conclusion. Start directly with the answer."
- "Only use information from the provided text. Do not add external knowledge."
- "Do not use bullet points. Write in paragraph form."
- "Respond in exactly 3 sentences."
Pro tip: Combine constraints with CREATE. The more specific your boundaries, the less the AI has to guess, and the more consistent your outputs become.
Concrete Examples
See the transformation from vague, ineffective prompts to structured, high-quality ones. Click "Show Analysis" to understand why.
Problem with the bad prompt: It lacks context (what meeting?), audience (who is it to?), tone (formal? casual?), and specific details (what happened?). The AI has to guess everything, leading to generic output.
Why the good prompt works: It uses CREATE — Role (project manager), Audience (client Sarah), Explicit Instructions (summarize + 3 action items), Tone (polite, encouraging), and Format (bullet points). Every element is specified.
Problem without CoT: LLMs can rush math and make arithmetic errors when they jump straight to the answer, especially with multi-step calculations involving conditions (the discount threshold).
Why CoT works: Forcing the model to lay out its intermediate steps (multiply, check threshold, calculate discount, subtract) drastically reduces calculation errors. Each step is simple enough to be accurate.
Problem with the vague request: "Nice" is subjective. The AI might be too casual, too formal, or miss the empathetic tone you actually want.
Why few-shot works: The two examples teach the AI the exact tone (apologetic, empathetic, action-oriented), structure (acknowledge, apologize, act), and level of formality. The third input gets transformed following the same pattern consistently.
Prompt Builder
Use the CREATE framework to build structured prompts. Try a preset or fill in the fields yourself.
Your Prompt Library
Saved prompts are stored locally in your browser. Export them anytime.
Prompt Engineering Quiz
Test your understanding of the concepts covered. Your best score is saved locally.
Comments
Post a Comment