AI Architecture Design

• ✓Turn a business problem into a scoped architecture with measurable quality targets
• ✓Design and communicate systems with C4 diagrams and Architecture Decision Records
• ✓Choose between API and self-hosted LLMs with evidence, not leaderboards
• ✓Design RAG pipelines for real documents, Arabic and English, and access control
• ✓Decide when an agent is justified, and design its tools, memory, approvals and guardrails
• ✓Threat-model AI systems: prompt injection, data leakage, tool misuse
• ✓Apply regional compliance (Saudi PDPL, Egypt's Data Protection Law, UAE PDPL)
• ✓Build evaluation and monitoring plans with release gates and rollback criteria
• ✓Learn the real trade-offs behind enterprise AI systems in production

Prerequisites

• 2+ years building software or ML systems in a team
• At least one service, API or model shipped to real users
• Comfortable with Python, REST APIs and basic cloud concepts (containers, managed databases)
• Hands-on exposure to LLMs (API calls, prompting, or a simple RAG)

1. Module 1

   Architecture Views, Steps and Process

   Sun 18 Oct. Frame the system before you design it.

2. Module 2

   AI System Building Blocks

   Sun 25 Oct. From data pipelines to serving, and what it costs.

3. Module 3

   LLM & RAG Architecture

   Sun 1 Nov. Choosing models with evidence and designing retrieval that works. Followed by the midpoint break (8–14 Nov): Midpoint Quiz and Capstone Part 1, no live session.

4. Module 4

   Agentic AI Architecture

   Sun 15 Nov. Agents that earn their place, and can't do damage.

5. Module 5

   Security, Governance, Evals

   Sun 22 Nov. Prove it's safe to ship, and know when it breaks.

6. Module 6

   Enterprise Case Studies

   Sun 29 Nov. How real systems were designed, and what we'd change. Capstone due Sun 13 Dec.

Times are shown in your timezone (Africa/Cairo).

1. 1.Session 1: Architecture Views, Steps and ProcessSun 18 Oct, 19:00 EEST · 90 min
   Topics

   Frame the system before you design it. • Introduction to architecture design for AI systems • Architecture viewpoints and stakeholders • Quality attributes: latency, cost, accuracy, security, availability • Design steps and verification • Frameworks: C4, TOGAF, Zachman • Real-world use case Session quiz (required). Optional assignment: choose your scenario, write the problem statement, rank quality attributes, and draw C4 Level 1 & 2.

2. 2.Session 2: AI System Building BlocksSun 25 Oct, 19:00 EEST · 90 min
   Topics

   From data pipelines to serving, and what it costs. • Data pipelines and feature stores • The model layer: foundation, fine-tuned and classical ML • Training vs inference infrastructure • Serving, APIs and orchestration • MLOps / LLMOps and observability • Real-world use case Session quiz (required). Optional assignment: C4 Level 3 component diagram, capability-to-model map, compute plan and monthly cost estimate.

3. 3.Session 3: LLM & RAG ArchitectureSun 1 Nov, 19:00 EET · 90 min
   Topics

   Choosing models with evidence and designing retrieval that works. • LLM selection and hosting: API vs self-hosted • Prompting and context management • Embeddings and vector databases • RAG pipelines: chunking, retrieval, re-ranking • Advanced patterns: hybrid search, GraphRAG • Real-world use case Session quiz (required). Optional assignment: model selection matrix, RAG pipeline diagram, retrieval strategy and context budget.

4. 4.Session 4: Agentic AI ArchitectureSun 15 Nov, 19:00 EET · 90 min
   Topics

   Agents that earn their place, and can't do damage. • Agent fundamentals and patterns: ReAct, planner–executor, router • Tools, function calling and MCP • Memory and state management • Multi-agent orchestration • Human-in-the-loop design • Guardrails and limits • Real-world use case Session quiz (required). Optional assignment: agent justification table, sequence diagram, tool catalog, memory and approval design.

5. 5.Session 5: Security, Governance, EvalsSun 22 Nov, 19:00 EET · 90 min
   Topics

   Prove it's safe to ship, and know when it breaks. • AI threat modelling: prompt injection, data leakage, tool misuse • Guardrails and access control • Governance, compliance and responsible AI in the MENA region • Evaluation methods: offline test sets, LLM-as-judge, human review • Production monitoring and observability • Real-world use case Session quiz (required). Optional assignment: threat model, compliance section, evaluation plan with release gates, monitoring plan.

6. 6.Session 6: Enterprise Case StudiesSun 29 Nov, 19:00 EET · 90 min
   Topics

   How real systems were designed, and what we'd change. • End-to-end architecture walkthroughs: telecom, e-commerce, government • The hardest trade-offs and how they were decided • What breaks at 10x scale • Lessons learned from production • Capstone guidance and Q&A Session quiz (required).

• E

  Eng. Sameh Amin

  Head of Artificial Intelligence, Unifonic

  Sameh has over 20 years in software and AI engineering and nearly a decade building AI organisations and production AI systems across the Middle East and Europe. At Unifonic, he founded the AI department from scratch. He launched an Arabic/English NLP engine, built a real-time SMS classifier processing 5,000 messages per second, and scaled AI to 6 squads of 20+ engineers delivering RAG, summarisation and text-generation services. He now leads Unifonic's agentic AI startup, which lets customers build working agents in under an hour. Before Unifonic, at DocuSign he designed an AI contract-negotiation system processing 100K+ contracts. At Etisalat he delivered the company's first NLP chatbot, a recommendation engine that lifted digital engagement by 70%, OCR document automation, and deep-learning face login. He holds an MSc in Artificial Intelligence from the University of Colorado Boulder and engineering leadership training from Cornell University.

• ✓Live sessions on Google Meet
• ✓Recordings on YouTube
• ✓Assignments
• ✓Final project
• ✓Certificate