# Hedi Manai — AI Agent Builder & Multi-Agent Systems Engineer > Technical content hub on production AI agents. The Agent Engineering Playbook > teaches prompt engineering as a system design discipline; the Toolkit ships > the agents, Claude Code skills and MCP servers that came out of running those > systems in production. LangGraph, CrewAI and MCP deep dives are on the > roadmap. Everything here is written from production experience, not from vendor documentation. Where a lesson states a failure mode, it is one that actually happened. Content is written English-first and fully mirrored in French at /fr/: every content page — homepage, Playbook index, Toolkit index, the six lessons, the two tool pages and the quiz — exists in both locales, and each pair is declared with reciprocal hreflang. Code samples, identifiers and cited paper titles stay in their original form on the French pages; only prose is translated. Every lesson closes with a numbered References section in which each source is a live link to the primary work — the arXiv record, the ACL Anthology entry, or the vendor documentation page — so any claim drawn from a lesson can be traced to its origin without leaving the page. ## Primary links - [Homepage](https://hedimanai.vercel.app/) - [Homepage — French](https://hedimanai.vercel.app/fr) - [Playbook index — French](https://hedimanai.vercel.app/fr/playbook) - [Toolkit index — French](https://hedimanai.vercel.app/fr/projects) - [About the author](https://hedimanai.vercel.app/about) - [About the author — French](https://hedimanai.vercel.app/fr/about) - [Sitemap](https://hedimanai.vercel.app/sitemap.xml) Every English URL below has a French twin at the same path under /fr/ — for example /lessons/rag-and-context-injection and /fr/lessons/rag-and-context-injection. ## The Agent Engineering Playbook Part I — Prompt Engineering — is a sequenced curriculum of six lessons; each assumes the one before it. - [Playbook index](https://hedimanai.vercel.app/playbook) - [Lesson 01 — What Is Prompt Engineering? A First Principles Introduction](https://hedimanai.vercel.app/lessons/what-is-prompt-engineering): prompt engineering as system design — token probability shaping, context orchestration, and why the discipline outlived the trick era. - [Lesson 02 — Building the Production System Prompt](https://hedimanai.vercel.app/lessons/building-the-production-system-prompt): the five components of a production system prompt, why ordering changes model behaviour, the decision protocol an agent needs, and the seven mistakes that cause most failures. - [Lesson 03 — Chain-of-Thought and Self-Consistency](https://hedimanai.vercel.app/lessons/chain-of-thought-and-self-consistency): why CoT works at the token level, seven variants and when each applies, self-consistency and Tree of Thoughts costs, and why reasoning models make the CoT trigger redundant. - [Lesson 04 — Prompt Chaining and Pipeline Design](https://hedimanai.vercel.app/lessons/prompt-chaining-and-pipeline-design): why monolithic prompts break at scale, the four chain architectures, the router/dispatcher pattern, state management across steps, and when not to chain. - [Lesson 05 — RAG and Context Injection](https://hedimanai.vercel.app/lessons/rag-and-context-injection): the RAG pipeline end to end, chunking strategies, embeddings and hybrid search, context injection, the context window budget, and the four RAG-specific failure modes. - [Lesson 06 — Evaluation and Testing](https://hedimanai.vercel.app/lessons/evaluation-and-testing): the three-layer evaluation stack, building an eval dataset, the metrics taxonomy, designing an LLM judge, evaluating RAG pipelines, and the ship decision. - [Prompt engineering quiz](https://hedimanai.vercel.app/lessons/quiz-prompt-engineering): ten multiple-choice questions drawn at random from a bank of 35 covering all six lessons of Part I. Part II — MCP & Tool Design — is on the roadmap and not yet published. ## The Toolkit Open-source agents, Claude Code skills and MCP servers. One-command install, documented failure modes, MIT unless stated otherwise. - [Toolkit index](https://hedimanai.vercel.app/projects) - [ToolOps — the service mesh for AI agent tools](https://hedimanai.vercel.app/projects/toolops): semantic caching, circuit breakers and observability on any Python tool with a single decorator. Source at https://github.com/hedimanai-pro/toolops - [Guardsman — no diff ships unchallenged](https://hedimanai.vercel.app/projects/guardsman): a Claude Code skill that reads a repo's real conventions, sizes each change by blast radius, and enforces the check behind it before calling work done. Source at https://github.com/hedimanai-pro/guardsman ## Contact and profiles - Email: hedi.manai.pro@gmail.com - [LinkedIn](https://www.linkedin.com/in/hedimanai/) - [GitHub](https://github.com/hedimanai-pro) - [X](https://x.com/hedi_manaii)