Prompt Engineering
Prompt Engineering is the practice of designing the input to an LLM to get desired outputs. Even with the most powerful model, a bad prompt produces terrible output.
Core techniques
1. Role Prompting
"You are a senior Python engineer specializing in code review..." typically yields more professional output than a bare question.
2. Few-shot Examples
Providing 2-3 input-output examples is much more effective than instructions alone.
3. Chain-of-Thought
"Please think step by step" or explicitly asking to show reasoning significantly improves complex task accuracy.
4. Structured Output
Explicitly request JSON / Markdown / table output for easy programmatic parsing.
5. Separators and Format
Use """ / <tag> / Markdown headers to clearly separate different parts
(system / context / question).
Advanced techniques
- Self-Consistency: sample multiple times, take majority answer.
- ReAct: interleaving Reasoning and Acting.
- Automatic Prompt Engineering: use models to optimize prompts (APE).
- Function Calling: structured tool invocation (see function-calling).
Debugging tools
- LangSmith / LangFuse: trace prompt call chains.
- PromptFoo: batch-run prompt test cases.
- PromptLayer: log every prompt and output.