Prompt Engineering Capstone

Your Prompt Engineering Capstone

This capstone brings together the major prompting skills covered throughout Module 5.

The goal is not simply to write a long prompt. The goal is to design a prompt that solves a real problem, produces a useful result, and can be improved through testing.

What You Have Learned

Throughout this module, you have learned how to:

• Understand prompt engineering.
• Define a clear role and goal.
• Add relevant context and background.
• Give clear instructions.
• Specify output formats.
• Use constraints effectively.
• Use examples to guide responses.
• Structure complex tasks.
• Ask AI to analyze and compare.
• Iterate and improve prompts.
• Prompt for summaries and transformations.
• Prompt for creative work.
• Prompt for research and analysis.
• Avoid common prompting mistakes.
• Use advanced prompting strategies.
• Build reusable prompt templates.
• Apply prompting to real-world tasks.

The Capstone Challenge

Choose one meaningful real-world task that you would genuinely like AI to help you complete.

Your task can involve:

• Writing.
• Research.
• Analysis.
• Planning.
• Communication.
• Learning.
• Business work.
• Customer service.
• Data interpretation.
• Creative work.

Step 1: Define the Problem

Begin by describing the task in plain language.

Ask:

• What am I trying to accomplish?
• Why is this task important?
• Who will use the result?
• What would make the result successful?

Step 2: Define the Goal

Turn the problem into a clear AI task.

For example:

Help me create a concise monthly management report from the supplied operational data.

Step 3: Define the Audience

Identify who will read or use the result.

For example:

The audience is senior managers who need a concise overview and do not need technical details.

Step 4: Provide Context

Give AI the background information that materially affects the task.

Do not add irrelevant information simply to make the prompt longer.

Step 5: Define the Input

Specify what information AI will receive.

For example:

The input will be a monthly spreadsheet containing sales, costs, customer activity, and operational metrics.

Step 6: Define the Instructions

Tell AI what it should do with the information.

1. Identify significant changes.
2. Identify unusual values.
3. Summarize the major trends.
4. Identify potential concerns.
5. Provide practical observations.

Step 7: Define the Output

Specify how the final result should be presented.

For example:

Return the result using the sections: Executive Summary, Key Changes, Trends, Concerns, and Recommended Actions.

Step 8: Add Constraints

Identify important boundaries.

• Use only the supplied data.
• Do not invent missing values.
• Preserve numerical values exactly.
• Keep the executive summary under 150 words.

Step 9: Decide Whether Examples Are Needed

If the task requires a particular style or classification approach, examples may help.

If the task is already clear without examples, do not add them unnecessarily.

Step 10: Define How Missing Information Should Be Handled

Real-world inputs are often incomplete.

For example:

If required information is missing, identify the missing information rather than guessing.

Step 11: Define How Uncertainty Should Be Handled

Tell AI to distinguish strong conclusions from assumptions or uncertain interpretations.

For example:

Clearly identify conclusions that depend on incomplete information or assumptions.

Step 12: Create the First Version

Combine the requirements into one complete prompt.

A capstone prompt might look like this:

You are assisting a senior management team. Analyze the supplied monthly operational data and identify the most significant changes, trends, unusual values, and potential concerns. Focus on information that could affect management decisions. Use only the supplied data and do not invent missing values. Preserve numerical values exactly as provided. Clearly distinguish observations from interpretations. If information is insufficient to support a conclusion, state that explicitly. Return the result using these sections: Executive Summary, Key Changes, Trends, Concerns, and Recommended Actions. Keep the Executive Summary under 150 words.

Step 13: Test the Prompt

Run the prompt with realistic input.

Do not evaluate the prompt only by how sophisticated it looks. Evaluate the actual result.

Step 14: Evaluate the Result

Compare the response with the original requirements.

• Did AI perform the requested task?
• Did it understand the context?
• Did it follow the output structure?
• Did it respect the constraints?
• Did it avoid inventing information?
• Did it provide useful detail?
• Was anything important missing?

Step 15: Identify Weaknesses

Do not simply say that the answer is not good enough.

Identify specific problems.

• The summary is too long.
• The response does not distinguish facts from interpretations.
• The recommendations are too generic.
• The output does not follow the requested structure.
• The AI made assumptions about missing data.

Step 16: Refine the Prompt

Modify the prompt to address the specific weaknesses.

For example:

For every recommendation, identify the finding that supports it. Do not provide a recommendation when the supplied information is insufficient to support one.

Step 17: Test Again

Run the revised prompt using the same input when possible.

Compare the new result with the first result.

Step 18: Evaluate the Improvement

Ask whether the change actually solved the problem.

If it did not, identify why and refine the prompt again.

Step 19: Create the Reusable Version

If the task is likely to occur again, convert the successful prompt into a reusable template.

For example:

Analyze the supplied [REPORT TYPE] for [AUDIENCE]. Focus on [FOCUS AREAS]. Use only the supplied information. Identify [REQUIRED FINDINGS]. Present the result using [OUTPUT STRUCTURE]. If information is missing, state that clearly rather than guessing.

Step 20: Document the Template

Record:

• Template name.
• Purpose.
• Required inputs.
• Optional inputs.
• Expected output.
• Important constraints.
• Known limitations.

Capstone Example: Customer Feedback

Suppose the task is to analyze customer feedback.

A complete prompt could be:

Analyze the supplied customer comments for a customer service manager. Classify each comment into one primary category, identify the five most common issues, provide evidence for each issue, and recommend practical actions. Use only the supplied comments. Do not infer customer information that is not present. If the evidence is insufficient to establish a likely cause, state that clearly. Return the result as: Summary, Category Analysis, Top Issues, Evidence, Recommendations, and Limitations.

Capstone Example: Research

A research workflow could use:

1. Define the research question.
2. Identify the scope.
3. Gather or provide relevant information.
4. Organize the evidence.
5. Compare perspectives.
6. Identify uncertainties.
7. Develop conclusions.
8. Present the findings.

Capstone Example: Writing

A writing workflow could use:

1. Define the audience.
2. Define the purpose.
3. Specify the tone.
4. Provide the source information.
5. Define the structure.
6. Generate a draft.
7. Critique the draft.
8. Produce the final version.

Capstone Example: Decision Support

A decision-support workflow could use:

1. Define the decision.
2. List the available options.
3. Define evaluation criteria.
4. Compare the options.
5. Identify trade-offs.
6. Identify risks and assumptions.
7. Identify missing information.
8. Provide a recommendation only after the analysis.

Capstone Example: Learning

A learning workflow could use:

1. Define the learner level.
2. Define the learning objective.
3. Explain the concept.
4. Provide examples.
5. Ask practice questions.
6. Evaluate the learner responses.
7. Adjust the explanation based on weaknesses.

Evaluate the Whole Workflow

The capstone is not only about the final answer.

Evaluate:

• The original prompt.
• The input provided.
• The intermediate results.
• The revised prompt.
• The final result.
• The verification process.

Human Review

Even a carefully designed prompt does not eliminate the need for human judgment.

Review the result for accuracy, relevance, completeness, and suitability.

Verification

Important factual claims should be checked against appropriate sources or source material.

Prompt engineering improves how AI performs a task; it does not guarantee that every generated statement is correct.

Know the Limits of the Capstone

A successful prompt does not mean AI should automatically make the final decision.

For high-impact tasks, human expertise and appropriate external tools may still be required.

Your Final Prompt Engineering Checklist

• Is the task clearly defined?
• Is the goal clear?
• Is the audience identified?
• Is relevant context provided?
• Is the input clearly identified?
• Are the instructions specific?
• Are important constraints included?
• Is the output format clear?
• Are examples useful for this task?
• Is missing information handled explicitly?
• Is uncertainty handled appropriately?
• Are the instructions consistent?
• Has the prompt been tested?
• Has the output been evaluated?
• Has the prompt been refined?
• Has important information been verified?

What You Have Accomplished

By completing this capstone, you have moved from simply asking AI questions to deliberately designing AI-assisted workflows.

You can now define tasks, communicate requirements, structure complex work, evaluate results, refine prompts, and create reusable prompting patterns.

Module 5 Complete

This capstone concludes the Prompt Engineering module.

The next module moves from prompting techniques to the broader landscape of AI Tools and how to choose and use different AI applications for different tasks.