prompt-engineering-patterns

Improve prompt consistency by adding specific constraints, including 2-3 representative few-shot examples, and using self-consistency sampling across multiple reasoning paths. Track accuracy and reproducibility metrics across diverse inputs to measure improvements.

Chain-of-thought prompting elicits step-by-step reasoning from the model, either zero-shot with phrases like "Let's think step by step" or few-shot with reasoning traces. Use it for multi-step reasoning tasks like math, logic, or complex analysis.

Balance example count against context window limits, typically starting with 2-3 examples selected by semantic similarity to the input. Add more only when accuracy gains justify the additional token cost and latency.

Reduce token usage by removing redundant phrases, consolidating similar instructions, using consistent abbreviations, and moving stable content into system prompts. Measure quality metrics after each change to confirm no degradation.

Inconsistent outputs usually come from vague instructions, ambiguous wording, or examples that do not match the target task. Add explicit output format specifications, test on edge cases, and include fallback instructions for uncertain inputs.

Avoid complex prompts before trying simple direct instructions, since over-engineering adds tokens and latency without guaranteed gains. Follow progressive disclosure: start simple, then add constraints, reasoning, and examples only as needed.