Master Generative AI (Artificial Intelligence) Course

Lesson 1 • What Makes AI Generative

Distinguishes generative from discriminative models and explains the core objective of learning data distributions. Positions generative AI as a distinct paradigm.

Lesson 2 • Data Representation and Preprocessing

Explains how text, images, and audio are encoded for neural networks. Proper data preparation directly affects generative model quality.

Lesson 3 • AI and Machine Learning Basics

Covers the hierarchy from AI to deep learning and how supervised, unsupervised, and reinforcement learning differ. Establishes vocabulary used throughout the course.

Lesson 4 • Neural Network Fundamentals

Introduces perceptrons, activation functions, backpropagation, and gradient descent. Provides the mathematical intuition needed for understanding generative architectures.

Lesson 1 • Comparing and Selecting Architectures

Provides a structured framework for evaluating architectures by quality, diversity, speed, and controllability. Prepares students to make informed design decisions.

Lesson 2 • Generative Adversarial Networks

Covers the generator-discriminator game, training dynamics, and common failure modes like mode collapse. Builds intuition for adversarial training used in image synthesis.

Lesson 3 • Variational Autoencoders

Explains the encoder-decoder structure, the reparameterisation trick, and the ELBO loss. Connects VAE theory to practical image and data generation tasks.

Lesson 4 • Diffusion Models

Introduces forward noising and reverse denoising processes, score matching, and DDPM. Explains why diffusion models achieve state-of-the-art image quality.

Lesson 1 • Tokenisation and Vocabulary Design

Explains byte-pair encoding, WordPiece, and SentencePiece and their effect on model performance. Tokenisation choices directly affect multilingual and code generation quality.

Lesson 2 • Scaling Laws and Emergent Abilities

Examines how model size, data, and compute interact to produce predictable and emergent behaviours. Helps students reason about capability thresholds.

Lesson 3 • LLM Failure Modes and Limitations

Analyses hallucination, sycophancy, context window limits, and knowledge cutoffs. Understanding failures is essential before applying LLMs in production.

Lesson 4 • Transformer Architecture Explained

Breaks down self-attention, multi-head attention, positional encoding, and feed-forward layers. This architecture underpins every major LLM covered later.

Lesson 1 • Advanced Prompting Strategies

Explores tree-of-thought, ReAct, and meta-prompting frameworks for multi-step problem solving. Extends basic prompting to agentic and iterative workflows.

Lesson 2 • Image and Multimodal Prompt Techniques

Covers text-to-image prompt anatomy, negative prompts, style tokens, and vision-language model inputs. Extends prompt skills beyond text-only models.

Lesson 3 • Prompt Evaluation and Testing

Introduces systematic prompt testing, regression suites, and automated evaluation metrics. Ensures prompt quality is measurable and reproducible.

Lesson 4 • Few-Shot and Chain-of-Thought Prompting

Covers in-context learning with examples and step-by-step reasoning elicitation. These techniques significantly improve accuracy on complex reasoning tasks.

Lesson 1 • Instruction Tuning and Alignment

Covers supervised instruction tuning, RLHF, and DPO for aligning model outputs with human preferences. Directly relevant to building safe, helpful AI assistants.

Lesson 2 • Domain Adaptation Strategies

Explains continued pre-training, domain-specific vocabulary extension, and data mixing ratios. Prepares students to adapt general models to specialised fields.

Lesson 3 • Fine-Tuning for Image Generation

Covers DreamBooth, textual inversion, and LoRA for diffusion models to personalise image generation. Extends fine-tuning skills to visual generative models.

Lesson 4 • Full Fine-Tuning Fundamentals

Covers dataset preparation, learning rate scheduling, and catastrophic forgetting mitigation for full model updates. Establishes the baseline before parameter-efficient methods.

Lesson 1 • Building RAG Pipelines

Explains document chunking, retrieval strategies, context injection, and answer generation. Connects retrieval components into an end-to-end grounded generation system.

Lesson 2 • Multi-Agent Systems and Orchestration

Covers agent roles, communication protocols, and orchestration frameworks for coordinating multiple agents. Enables students to build complex automated workflows.

Lesson 3 • Vector Databases and Embeddings

Covers dense vector embeddings, similarity search algorithms, and vector database indexing. These components form the retrieval backbone of RAG systems.

Lesson 4 • AI Agent Fundamentals

Introduces the agent loop, tool use, memory types, and planning strategies. Establishes the conceptual model for autonomous AI systems.

Lesson 1 • Human Evaluation Design

Explains annotation task design, inter-annotator agreement, and crowdsourcing best practices. Human evaluation remains the gold standard for nuanced quality assessment.

Lesson 2 • Image and Multimodal Evaluation

Covers FID, CLIP score, and human preference studies for evaluating image and multimodal outputs. Extends evaluation skills beyond text-only systems.

Lesson 3 • Safety and Bias Evaluation

Introduces red-teaming, toxicity benchmarks, and fairness audits for generative models. Ensures evaluation covers harms, not just quality.

Lesson 4 • Automatic Metrics for Text Generation

Covers BLEU, ROUGE, BERTScore, and perplexity and explains when each metric is appropriate. Provides a toolkit for automated quality measurement.

Lesson 1 • Model Optimisation for Inference

Covers quantisation, pruning, distillation, and speculative decoding to reduce latency and cost. Optimisation is essential before any production deployment.

Lesson 2 • MLOps for Generative AI

Covers versioning, CI/CD pipelines, experiment tracking, and model registries adapted for generative workloads. Operationalises the full model lifecycle.

Lesson 3 • Guardrails and Output Filtering

Covers input validation, output classifiers, content moderation layers, and fallback strategies. Guardrails prevent harmful or off-policy outputs in live systems.

Lesson 4 • Monitoring and Observability

Introduces latency tracking, output drift detection, cost monitoring, and alerting pipelines. Ongoing observability ensures production systems remain reliable.