Video Curriculum // Structured Playlists

Learn it step by step.

Structured, module-by-module video courses on topics I work with every day: AI strategy for consultants, reinforcement learning in Python, graph-based retrieval, context engineering, agent memory, and distributed microservice architecture.

Each module is a self-contained 15–35 minute lesson on YouTube or GitHub. Tick off modules with the checkmark button on the right to track your progress locally in your browser.

6Complete Courses
49Video Modules
~15.7hTotal Runtime
0 / 53Lessons Completed
COURSE 01 · ORIENTATION
Technical orientation 5 modules ~1.5 hours November 2026

AI for Business Consultants: A Technical Orientation

AI for Business Consultants: A Technical Orientation course cover ▶

A panoramic, technically literate view of AI for the consultant who has to lead the room. It moves quickly through classical machine learning, spends its weight on generative AI, agents, platforms and guardrails, then turns to what makes AI transformations succeed or fail. The last module scores two real cases against that framework and ends with an interview toolkit. Every figure on screen carries its source, and the vendor landscape is dated November 2026.

Who it's for: business and technology consultants, and anyone who knows software and finance but not machine learning, and has to challenge vendors and engineers, lead AI-transformation engagements or sit an interview on them. It follows the book The AI Transformation Consultant.

Generative AILLMsAgentsMCPGuardrailsAI TransformationEU AI ActCase Studies

Course modules

  1. ▶ Course trailer 0:32
  2. Module 1AI in One Sitting: From Rules to Learning to TransformersThe mental-model shift from rules to learning, the AI map, classical machine learning, deep learning, model quality for non-engineers, MLOps, and why data is the real asset. 19:22
  3. Module 2Generative AI and LLMs, Inside OutHow an LLM works and is built, its limits, prompting and context, RAG, open versus closed models, the vendor landscape, and how to choose with your own evals. 17:38
  4. Module 3Agents, Platforms, Hyperscalers and GuardrailsAgents, tool use, MCP and A2A, frameworks, the hyperscalers and wider ecosystem, a reference architecture, guardrails, LLMOps, and the EU AI Act and NIST rules. 16:53
  5. Module 4Making AI Transformation WorkWhy transformations fail and what makes them succeed: the eight failure modes and seven success factors, a consulting playbook, business case and ROI, operating model and industry lenses. 16:51
  6. Module 5Case Studies and the Interview ToolkitA success (JPMorgan Chase) and a failure (MD Anderson with IBM Watson) scored against the framework, a lightning round of four more cases, 15 interview questions and a 30-day plan. 16:48
COURSE 02 · HANDS-ON PYTHON
Beginner → Intermediate 5 modules ~2.6 hours English + Egyptian Arabic October 2026

Reinforcement Learning for Python Developers with Stable-Baselines3

Reinforcement Learning for Python Developers using Stable-Baselines3, course cover with the course robot

Train agents that learn by trial and error, with no heavy math. Each module teaches one idea with a memorable analogy, then runs short, real Stable-Baselines3 code and shows the measured result: a dog trainer for PPO, a restaurant critic's notebook for DQN, a steady hand for SAC and TD3. One project, a CoffeeShop that orders beans, grows through all five modules, and the last module ships it. 34 runnable examples, every number on screen comes from a real run.

Who it's for: Python developers who know classes and pip and want to use reinforcement learning. No calculus, no statistics. The course is available in English and in Egyptian Arabic (Arabic narration, English terms, code and on-screen text).

بالعربي: الكورس متاح بنسخة بالعربي المصري: شرح بالمصري، والـ code والمصطلحات بالإنجليزي، في خمس أجزاء.

PythonReinforcement LearningStable-Baselines3PPODQNSACGymnasiumPyTorch

Videos: coming soon to YouTube. The code, notes and examples are public now; run times below are the English videos (the Arabic videos run about 10% shorter).

Course modules

  1. Module 1The Game LoopAgent, environment and reward as plain Python objects, the reset() / step() contract, spaces, a first trained player in one line, and your own CoffeeShop environment. 24:54
  2. Module 2The Dog Trainer (PPO and A2C)Why PPO is the safe default, the trainer knobs (n_steps, gamma, clip_range), parallel environments, TensorBoard, and a fair PPO vs A2C race. 32:22
  3. Module 3The Notebook of Experience (DQN)Q-values as a notebook, the replay buffer, epsilon-greedy exploration, the frozen target network, and why DQN needs buttons, not dials. 29:55
  4. Module 4The Steady Hand (SAC, TD3, DDPG)Continuous actions, entropy as curiosity, twin critics and exploration noise: a three-way race on the Pendulum, then CoffeeShop with a dial. 31:59
  5. Module 5The Workshop (shipping it)Reward hacking, callbacks, save and load, VecNormalize, CPU vs GPU, Atari pixels, the algorithm chooser, and shipping your own bot. 36:55
COURSE 03 · ARCHITECTURE
Beginner → Intermediate 4 modules ~1.3 hours September 2026

A Gentle Introduction to Microservices for Python Developers

A Gentle Introduction to Microservices for Python Developers course cover ▶

You can build a web app in Python, but you've never designed a distributed system. This is a high-level, architect's-eye course on how large applications are designed, decomposed, built and deployed as microservices on AWS and Google Cloud. It follows the book A Gentle Introduction to Microservices for Python Developers, one module per major part of the book.

Who it's for: Python developers and aspiring architects who want the big picture before diving into frameworks.

PythonMicroservicesAWSGoogle CloudArchitecture

Course modules

  1. ▶ Course trailer 0:11
  2. Module 1Microservices FundamentalsWhat microservices are, how they differ from a monolith, and when the trade-offs are worth it. 18:07
  3. Module 2Designing & Decomposing a Large SystemHow to break a big system into services with clear boundaries and responsibilities. 16:58
  4. Module 3Microservice Architecture PatternsThe recurring patterns that make a set of services work together reliably. 19:27
  5. Module 4Building Microservices on AWS & Google CloudTaking the design to the cloud: building and deploying services on AWS and Google Cloud. 23:30
COURSE 04 · ADVANCED SERIES I
Advanced 8 modules ~2.5 hours June 2026

Advanced GraphRAG & Knowledge Agents

Advanced GraphRAG & Knowledge Agents course cover ▶

A friendly, in-depth course on building structured, graph-based retrieval systems, from first principles to production multi-agent pipelines. It starts with why plain vector RAG hits a ceiling, then walks through the major GraphRAG families: ontology-driven graph construction, Microsoft GraphRAG and DRIFT, LightRAG, HippoRAG, LinearRAG, AGRAG and StructRAG, finishing with multi-tiered architectures and benchmarks.

Who it's for: Engineers and data scientists who already know basic RAG and want to go deeper into graphs.

GraphRAGKnowledge GraphsRAGNeo4jLightRAGHippoRAG

Course modules

  1. ▶ Series intro 0:27
  2. Module 1The RAG Bottleneck & Evolution of Graph-Based RetrievalWhere flat vector retrieval breaks down, and how graph-based retrieval evolved to fix it. 13:08
  3. Module 2Ontological Foundations & Graph ConstructionTurning raw text into a knowledge graph with LlamaIndex and Neo4j. 13:47
  4. Module 3Hierarchical Community Indexing & Global-Local SearchMicrosoft GraphRAG and DRIFT: community detection, summaries, and global versus local queries. 18:54
  5. Module 4Dual-Level Retrieval & Incremental UpdatesLightRAG: dual-level retrieval and updating a graph without rebuilding it. 16:51
  6. Module 5Neurobiologically Inspired Long-Term MemoryHippoRAG: hippocampus-inspired indexing and Personalized PageRank retrieval. 20:35
  7. Module 6Relation-Free Hierarchical Graph ArchitecturesLinearRAG: building a graph without extracting relations. 19:53
  8. Module 7Minimum Cost Subgraphs & Inference-Time StructuringAGRAG and StructRAG: picking the cheapest useful subgraph and structuring knowledge at query time. 20:14
  9. Module 8Multi-Tiered Architectures, Workflows & BenchmarksPutting the pieces together into multi-tier pipelines and measuring them with GraphRAG-Bench. 25:12
COURSE 05 · ADVANCED SERIES II
Advanced · Sequel 11 modules ~3.8 hours August – September 2026

Advanced GraphRAG & Knowledge Agents — Part 2

Advanced GraphRAG & Knowledge Agents — Part 2 course cover ▶

The sequel to Part 1 covers the next five families of graph retrieval. It begins with LazyGraphRAG, which defers the cost of indexing, then spends a long module on temporal and agent-memory graphs (two clocks, validity intervals, contradiction handling, Graphiti and Zep, and the wider memory-system landscape). It closes with learned graph retrievers, hypergraphs and agentic graph traversal.

Who it's for: Anyone who finished Part 1, or who is building agents that need memory that changes over time.

Agent MemoryTemporal GraphsGraphitiMem0HypergraphsGNN-RAG

Course modules

  1. ▶ Part II intro 0:29
  2. Module 1LazyGraphRAG: Deferring the Cost of UnderstandingBuild a usable graph with zero LLM calls at index time and pay for understanding only when a question arrives. 24:34
  3. Module 2AWhy Static Memory BreaksReal agents accumulate facts that change. Why a graph that assumes a still corpus ends up confidently wrong. 13:07
  4. Module 2BTwo Clocks: Event Time vs Ingestion TimeEvery fact has two timestamps: when it happened and when you learned it. 15:12
  5. Module 2CValidity Intervals & As-Of QueriesEdge validity fields, and asking both what was true on a date and what you believed on it. 17:56
  6. Module 2DContradiction & InvalidationWhat to do when a new fact disagrees with a stored one, even when they share no words. 13:24
  7. Module 2EMeet Graphiti & ZepA system that implements the whole temporal-memory specification built up so far. 18:39
  8. Module 2FThe Memory Systems LandscapeMem0, A-MEM, G-Memory and ATOM: what each is and the one idea that makes it different. 16:39
  9. Module 2GAuditability & Choosing Your MemorySAGE and MOSS on trust and auditability, plus how to choose a memory system with no common leaderboard. 16:32
  10. Module 3Learned Graph RetrieversGNN-RAG and GFM-RAG: retrievers that are trained rather than computed in closed form. 25:49
  11. Module 4Beyond Triples: Hypergraphs, Atoms & TreesRepresenting facts that are not binary, with hypergraphs, atomic facts and tree structures. 28:44
  12. Module 5Agentic Graph TraversalThink-on-Graph, KAG and Structure-R1: retrievers that walk the graph, look, decide and look again. 36:17
COURSE 06 · DEEP RESEARCH TOUR
Beginner → Advanced 16 modules ~4.0 hours October 2026

Context Management & Agent Memory in LLMs

Context Management & Agent Memory in LLMs course cover ▶

Why does your AI assistant forget? This gentle, visual course starts with the basics (what context is and why models forget), then moves slowly through recent research on how agents manage context and memory: just-in-time retrieval, learned summaries, pruning, folding, memory editing, evolving playbooks, memory operating systems, graph memory and memory beyond text. Animated diagrams, slow and clear explanations, and real code demos run on local models.

Who it's for: Anyone building or using LLM agents who wants to understand context windows, long-task memory and the newest research, with no prior knowledge needed for the first two modules.

Context EngineeringAgent MemoryLLM AgentsSummarizationMemory GraphsLocal Models

Course modules

  1. ▶ Course introduction 0:50
  2. Module 1What Is Context?A gentle start: what "context" means for an AI model, how much of it the model can see at once, and why simply adding more is not always better. 11:03
  3. Module 2Why Models ForgetWhy long conversations and long agent tasks make models lose track: the ever-growing record, the "lost in the middle" effect, four ways context fails, and the difference between forgetting and memory. 12:03
  4. Module 3A Map of the FieldTwo maps of the field, agent memory and shrinking the context, so that every later module has a place to land. 13:49
  5. Module 4Four Production PatternsFour patterns used in real agent products to keep context under control: fetch just in time, clear old results, compact the history, and hand work to sub-agents. Plus how to choose between them. 11:22
  6. Module 5The Surprisingly Simple WinnerA surprising result from JetBrains Research (The Complexity Trap): on software-engineering tasks, a very simple context strategy holds its own against more elaborate ones. The three strategies, the results, why summaries fall short, a hybrid, and cautions. 12:38
  7. Module 6Summaries That LearnSummaries that are learned instead of hand-written: ReSum (summarize, then restart) and ACON (learning what to keep), with a hands-on look at the summary step and how the two compare. 14:36
  8. Module 7Skimming Tool OutputSWE-Pruner: a small helper that skims code the way a reader does and keeps only what matters for the goal, with a hands-on run. Then SWE-Pruner Pro, which needs no second model. 13:59
  9. Module 8Rebuilding the WorkspaceIterResearch: instead of piling everything onto the agent's desk, rebuild a compact workspace every round. How it is trained for efficiency, and the results. 13:23
  10. Module 9Folding the TrajectoryAgentFold and Context-Folding: letting the agent decide when to fold its past into a compact record, or branch off a sub-task and return, with a hands-on run of the real branch loop. 15:12
  11. Module 10Letting the Agent Edit Its Own MemoryTeaching agents to edit their own memory: Memory-as-Action, AgeMem with its two kinds of memory, and FoldGRPO, which trains the fold. 17:09
  12. Module 11Context as an Evolving PlaybookAgentic Context Engineering (ACE): treat context as an evolving playbook with three roles instead of rewriting it from scratch. The results and a hands-on run of the real code. 14:36
  13. Module 12Lifelong Memory, Far CheaperSimpleMem: a three-stage approach (write, tidy, look up) that gives agents lifelong memory at a much lower cost. The results and a hands-on run of the real code. 14:06
  14. Module 13A Memory Operating SystemEverMemOS: a self-organizing memory operating system built from MemCells, scenes and recollection. The results and a hands-on run of its successor code, EverOS. 18:42
  15. Module 14Memory as GraphsMAGMA: agent memory as several connected graphs, with four kinds of connection, retrieval paths chosen by the question, and two speeds of writing. The results and a hands-on run of the real code. 19:49
  16. Module 15Beyond Text: Latent and Parametric MemoryBeyond text: three places to keep memory. AgentOCR (history as a picture), CommitKV (trimming the model's notes) and Cortex (memory in the weights). 22:01
  17. Module 16Putting It All TogetherPutting it all together: a lifecycle for context, how memory is measured (LongMemEval and LoCoMo), a decision guide, and what to do next. 17:39

New modules are added to these playlists as they publish. Subscribe on YouTube ↗ to follow along.