ai-prompt-engineering-expert

# AI Prompt Engineering & Automated Evals Expert (2026 Edition) [English](#english) | [Bahasa Indonesia](#bahasa-indonesia) --- <a name="english"></a> ## English ### Description Production-grade guide covering prompt engineering and automated evaluation (Evals). Teaches how to write, version, defend, benchmark, and regression-test LLM prompts and agent workflows using **Promptfoo**, **DeepEval**, and structured JSON schemas. ### Trigger Conditions - Writing or refactoring system prompts for autonomous AI agents. - Enforcing strict structured output (JSON Schema / Zod). - Defending against Prompt Injection or jailbreak attacks. - Setting up automated regression testing and CI/CD quality gates for LLMs. - Benchmarking RAG output quality (Faithfulness, Relevance, Hallucinations). --- ### Part 1: Prompt Construction & Defense #### 1. Structured Output (Schema-First) Never rely on prompt instructions alone to get JSON. Always use native Tool Calling / Structured Outputs with JSON Schema or Zod: ```typescript import { z } from 'zod'; export const UserAnalysisSchema = z.object({ sentiment: z.enum(['positive', 'neutral', 'negative']), confidence: z.number().min(0).max(1), tags: z.array(z.string()), }); ``` #### 2. Advanced Prompting Techniques - **Chain-of-Thought (CoT)**: Direct the model to deliberate before producing final answers. Instruct output inside `<thinking>` tags. - **Few-Shot Prompting**: Provide 2-3 diverse input-output...