skip to content

Prompting

6 cheat sheets

Prompting#

Foundational prompt engineering patterns, RAG, evals, few-shot, chain-of-thought, and structured output — the techniques that make LLM behavior reliable, measurable, and production-ready regardless of which model or SDK you use.

What’s in this section#

PageTopics
Prompt PatternsRole prompts, prefilling, XML structure, system vs user, P-T-C-O, vision, prompt caching
RAG ChecklistIngestion, embedding, retrieval, vector DBs, hybrid search, observability, eval
EvaluationsGolden datasets, LLM-as-judge, rubrics, statistical significance, regression detection
Few-Shot PromptingExample selection, count tuning, dynamic retrieval, many-shot ICL, fine-tuning trade-off
Chain-of-ThoughtZero-shot / few-shot CoT, self-consistency, tree of thoughts, reasoning models
Structured OutputJSON mode, tool use as schema channel, Pydantic / Zod validation, retry loops

Sections#

  • Prompt Engineering Patterns — Repeatable techniques: role prompts, CoT, few-shot, XML structure, output schemas, vision, prefilling, and extended thinking.
  • RAG Implementation Checklist — End-to-end checklist covering document ingestion, chunking strategies, embedding models, retrieval tuning, hybrid search, and evaluation.
  • LLM Evaluations — Golden datasets, LLM-as-judge, rubrics, statistical significance, regression detection, evals vs tests.
  • Few-Shot Prompting — In-context learning fundamentals, example selection, format design, dynamic retrieval, and order effects.
  • Chain-of-Thought Prompting — Zero-shot and few-shot CoT, self-consistency, tree of thoughts, and reasoning-model comparison.
  • Structured Output — Tool use as a schema channel, Pydantic/Zod pairing, JSON mode, and validator retry loops.

Prompt Engineering Patterns

Reliable prompt structures for reasoning, extraction, classification, generation, extended thinking, and vision tasks with Claude.

May 25, 2026
prompting llm ai claude chain-of-thought xml thinking

RAG Implementation Checklist

End-to-end checklist and code for building reliable Retrieval-Augmented Generation pipelines — chunking, embedding, vector DBs, retrieval, and evaluation.

May 25, 2026
rag ai embeddings vector-db llm retrieval semantic-search

LLM Evaluations

Build production evaluation pipelines for LLM applications — golden datasets, LLM-as-judge, rubrics, statistical significance, regression detection, and evals vs tests.

May 25, 2026
evals evaluation llm-as-judge golden-dataset regression statistics prompting

Few-Shot Prompting

In-context learning techniques — example selection, format design, count tuning, dynamic retrieval of demonstrations, and pitfalls of few-shot prompting.

May 25, 2026
few-shot in-context-learning icl prompting llm examples demonstrations

Chain-of-Thought Prompting

CoT prompting techniques — zero-shot CoT, few-shot CoT, self-consistency, tree of thoughts, and how reasoning models compare with prompted reasoning.

May 25, 2026
chain-of-thought cot reasoning llm prompting self-consistency tree-of-thoughts thinking

Structured Output

Techniques for reliable structured generation — JSON mode, schema-constrained decoding, function/tool calls as output, and validator pairing with Pydantic or Zod.

May 25, 2026
structured-output json schema pydantic zod tool-use function-calling prompting