A practical tutorial on building LLM output guardrails: input classification, output filtering, layered checks, and the failure modes that catch first-time implementers off guard.