prompt-engineering

Prompt Patterns

Zero-Shot

Translate the following text to Thai: "Hello, how are you?"

Few-Shot

Examples:
Input: "Good morning" → Output: "สวัสดีตอนเช้า"
Input: "Thank you" → Output: "ขอบคุณ"
Input: "Goodbye" → Output: ?

Chain-of-Thought (CoT)

Let's solve this step by step:
1. First, identify the problem
2. Break down the components
3. Apply relevant formula
4. Calculate the answer
5. Verify the result

---

Prompting Best Practices

Structure

# Role
You are an expert Python developer.

# Context
We are building a REST API using FastAPI.

# Task
Create a user registration endpoint.

# Requirements
- Validate email format
- Hash password with bcrypt
- Return JWT token

# Output Format
Return complete, working code with comments.

Be Specific ✅

❌ "Write code for a button"
✅ "Create a React button component that:
   - Has primary and secondary variants
   - Accepts onClick handler
   - Shows loading spinner when isLoading is true
   - Uses Tailwind CSS for styling"

---

Advanced Techniques

Self-Consistency

Generate 3 different solutions, then pick the best one.

Tree-of-Thought

Consider multiple approaches:
Approach A: [describe]
Approach B: [describe]
Approach C: [describe]
Compare and select optimal solution.

RAG Pattern

Context: [Retrieved documents]
Question: [User query]
Answer based ONLY on the context provided.

---

Temperature Guide

| Temperature | Use Case | |-------------|----------| | 0.0 - 0.3 | Factual, code generation | | 0.4 - 0.7 | Balanced creativity | | 0.8 - 1.0 | Creative writing |

---

Checklist

• [ ] Define clear role/persona
• [ ] Provide sufficient context
• [ ] Be specific about output format
• [ ] Use examples when helpful
• [ ] Iterate and refine prompts