Prompt Engineering Fundamentals: From Basics to Production

Essential prompt engineering patterns. Structure, context management, and techniques that consistently produce better outputs.

Good prompts aren’t magic. They’re engineering.

Consistent patterns produce consistent results.

Core Principles

1. Be Specific

Bad: “Write about dogs” Good: “Write a 200-word article about the health benefits of walking dogs, targeting first-time dog owners”

2. Provide Context

Bad: “Fix this code” Good: “Fix this TypeScript function that should validate email addresses. It currently allows invalid formats like ‘test@’ through.”

3. Define Format

Bad: “List some options” Good: “List 5 options as a numbered list. Each item should have a title and one-sentence description.”

4. Set Constraints

Bad: “Keep it short” Good: “Maximum 3 paragraphs, each under 50 words”

The RISEN Framework

Role: Who should the AI be? Instructions: What should it do? Steps: How should it proceed? End goal: What’s the desired output? Narrowing: What constraints apply?

You are a senior software engineer (Role).

Review this code for security vulnerabilities (Instructions).

First, identify input validation issues. Then, check for injection risks. Finally, look for authentication bypasses (Steps).

Provide a prioritized list of issues with severity ratings and fixes (End goal).

Focus only on security, not style or performance (Narrowing).

Template Patterns

Few-Shot Learning

Convert measurements to metric:

Input: 5 feet
Output: 1.52 meters

Input: 10 pounds
Output: 4.54 kilograms

Input: 32 degrees Fahrenheit
Output: 0 degrees Celsius

Input: {{user_input}}
Output:

Chain of Thought

Solve this step by step:

Problem: {{problem}}

Step 1: Identify what we know
Step 2: Identify what we need to find
Step 3: Apply relevant formulas
Step 4: Calculate the answer
Step 5: Verify by checking

Show your work for each step.

Structured Output

Analyze this text and return JSON:

Text: {{input}}

Return exactly this structure:
{
  "sentiment": "positive" | "neutral" | "negative",
  "topics": ["string", ...],
  "summary": "One sentence summary",
  "confidence": 0.0-1.0
}

Common Mistakes

1. Vague instructions → Ambiguous outputs
2. No examples → Inconsistent format
3. Too many tasks → Incomplete execution
4. Missing constraints → Verbose or wrong-length responses
5. Assuming context → AI doesn’t know what you know

Testing Prompts

async function testPrompt(prompt: string, testCases: TestCase[]) {
  const results = await Promise.all(
    testCases.map(async (test) => {
      const output = await complete(prompt.replace('{{input}}', test.input));
      return {
        input: test.input,
        output,
        passed: test.validate(output)
      };
    })
  );
  
  const passRate = results.filter(r => r.passed).length / results.length;
  return { results, passRate };
}

Aim for 90%+ pass rate before production.

Prompt engineering is foundational. See PromptRequest for our prompt management system.