Agentic Design Patterns — بالمصري 🇪🇬

The 21 agentic AI design patterns from Antonio Gulli's book, explained in Egyptian Arabic — with framework-free, runnable Python examples.

Every pattern gets: a plain-Arabic explanation built on analogies, a Mermaid diagram of the mechanism, when-to-use guidance and trade-offs verified against the book, and a self-contained Python example that demonstrates the real mechanism with nothing but the OpenAI SDK — no LangChain, no CrewAI, no magic.

ليه بالمصري؟ لأن أنماط تصميم الـ AI Agents هي أساس أي نظام ذكي حقيقي، والمحتوى العربي التقني الجاد فيها شبه معدوم. الريبو ده شرح كامل للـ 21 pattern من كتاب Antonio Gulli بلغة بسيطة وأمثلة كود تشتغل فعلاً.

Quick Start

git clone https://github.com/Muhannad-Khaled/agentic-design-patterns.git
cd agentic-design-patterns
pip install -r requirements.txt
cp .env.example .env   # add your API key (OpenAI, Groq, Ollama — anything OpenAI-compatible)
python examples/01_prompt_chaining.py

The Patterns

#	Pattern	TL;DR	Guide	Code
1	Prompt Chaining	Split a complex task into a pipeline of focused, validated LLM calls	📖	🐍
2	Routing	Classify each request and dispatch it to the right specialist	📖	🐍
3	Parallelization	Run independent sub-tasks concurrently and merge the results	📖	🐍
4	Reflection	Producer/critic loop — draft, critique, refine until quality passes	📖	🐍
5	Tool Use	Let the model call external functions and APIs to act on the world	📖	🐍
6	Planning	Generate a step-by-step plan before executing, adapt as things change	📖	🐍
7	Multi-Agent Collaboration	A team of specialized agents coordinated by an orchestrator	📖	🐍
8	Memory Management	Short-term vs long-term memory — what to store, compress, and recall	📖	🐍
9	Learning & Adaptation	Improve behavior over time from feedback and outcomes	📖	🐍
10	Model Context Protocol (MCP)	The open standard for discovering and calling external tools	📖	🐍
11	Goal Setting & Monitoring	Define measurable goals (KPIs) and self-correct on drift	📖	🐍
12	Exception Handling & Recovery	Classify failures, retry with backoff, fall back, escalate	📖	🐍
13	Human-in-the-Loop	Pause for human approval on high-stakes or ambiguous decisions	📖	🐍
14	Knowledge Retrieval (RAG)	Ground answers in your own documents via retrieval	📖	🐍
15	Inter-Agent Communication (A2A)	Structured messaging protocols between agents	📖	🐍
16	Resource-Aware Optimization	Route easy tasks to cheap models, hard ones to strong models	📖	🐍
17	Reasoning Techniques	CoT, ToT, self-correction, debate — structured thinking before answering	📖	🐍
18	Guardrails / Safety Patterns	Screen inputs and outputs for harm, PII leaks, and prompt injection	📖	🐍
19	Evaluation & Monitoring	Golden test sets, quality gates, and drift detection in production	📖	🐍
20	Prioritization	Score and order tasks by value, urgency, effort, and dependencies	📖	🐍
21	Exploration & Discovery	Open-ended research — explore, cluster, dig deeper, synthesize	📖	🐍

Which Pattern Do I Need?

flowchart TD
    START{"What's your main challenge?"} --> BIG["Task too big for<br/>one prompt"]
    START --> VARIED["Many different<br/>request types"]
    START --> WORLD["Needs external<br/>data / actions"]
    START --> QUALITY["Output quality<br/>is critical"]
    START --> SCALE["Multiple specialists<br/>working together"]
    START --> PROD["Running safely<br/>in production"]

    BIG --> B1{"Steps depend<br/>on each other?"}
    B1 -- yes --> P01["01 Prompt Chaining"]
    B1 -- no --> P03["03 Parallelization"]
    BIG --> P06["06 Planning"]

    VARIED --> P02["02 Routing"]
    VARIED --> P16["16 Resource-Aware<br/>Optimization"]
    VARIED --> P20["20 Prioritization"]

    WORLD --> W1{"What kind?"}
    W1 -- "APIs & actions" --> P05["05 Tool Use"]
    W1 -- "standard tool protocol" --> P10["10 MCP"]
    W1 -- "your documents" --> P14["14 RAG"]

    QUALITY --> P04["04 Reflection"]
    QUALITY --> P17["17 Reasoning<br/>Techniques"]
    QUALITY --> P19["19 Evaluation &<br/>Monitoring"]

    SCALE --> P07["07 Multi-Agent<br/>Collaboration"]
    SCALE --> P15["15 A2A<br/>Communication"]
    SCALE --> P08["08 Memory<br/>Management"]

    PROD --> P12["12 Exception<br/>Handling"]
    PROD --> P13["13 Human-in-<br/>the-Loop"]
    PROD --> P18["18 Guardrails"]
    PROD --> P11["11 Goal Setting &<br/>Monitoring"]

Cross-cutting patterns you'll layer on top: 09 Learning & Adaptation (systems that improve with feedback) and 21 Exploration & Discovery (open-ended research agents).

Appendices

	Appendix	About
A	Advanced Prompting Techniques	The prompting toolbox underneath every pattern
B	From GUI to Real-World Environments	Agents that see screens and act in the physical world
C	Overview of Agentic Frameworks	LangChain, LangGraph, CrewAI, Google ADK & friends
D	Building an Agent with AgentSpace	UI-based agent building
E	AI Agents on the CLI	Claude Code, Gemini CLI & terminal-native agents
F	Under the Hood: Reasoning Engines	How the thinking actually happens
G	Coding Agents	Agents that write software

Repo Structure

├── patterns/      # 21 pattern guides (Egyptian Arabic, one per book chapter)
├── appendices/    # 7 appendix summaries
├── examples/      # 21 runnable Python scripts — plain OpenAI SDK, no frameworks
├── requirements.txt
└── .env.example   # works with OpenAI, Groq, Ollama, or any OpenAI-compatible API

About the Source

This repository is an original Egyptian-Arabic study guide of:

Antonio Gulli — Agentic Design Patterns: A Hands-On Guide to Building Intelligent Systems, Springer, 2025. 📚 Springer edition · Free official version (Google Docs) · 100% of the author's royalties go to Save the Children

It summarizes and explains the book's concepts in a different language and style — it does not reproduce the book's text. All example code here is written from scratch. If this material helps you, please support the original book.

License

Code (examples/): MIT
Content (patterns/, appendices/, this README): CC BY 4.0

Built by Muhannad Khaled · Contributions and corrections are welcome — open an issue or PR.

Muhannad-Khaled/Agentic-AI-Design-Patterns

Star growth

README