Generative AI Course

Lesson 1 • What Generative AI Actually Is

Defines generative AI by contrasting it with discriminative AI and traditional software. Establishes the vocabulary used throughout the course.

Lesson 2 • Core Concepts: Probability and Patterns

Introduces the probabilistic nature of generative models without requiring advanced maths. Connects pattern learning to output generation.

Lesson 3 • Brief History of AI and GenAI

Traces AI from rule-based systems to modern neural networks and large models. Provides historical context that explains why current architectures exist.

Lesson 4 • Ethical and Societal Context

Frames responsible AI use as a foundational mindset, not an afterthought. Covers bias, misinformation risk, and accountability from day one.

Lesson 1 • Context Windows and Memory Limits

Explains the practical implications of finite context windows for real tasks. Prepares learners to design prompts and workflows that respect these limits.

Lesson 2 • Pre-training and the Role of Data

Covers how LLMs are trained on massive corpora using next-token prediction. Explains why data quality and scale drive model capability.

Lesson 3 • Transformer Architecture Essentials

Explains attention mechanisms, encoder-decoder structures, and positional encoding at a conceptual level. Grounds later prompt engineering in architectural reality.

Lesson 4 • Fine-tuning and Alignment Techniques

Introduces supervised fine-tuning and reinforcement learning from human feedback. Shows how raw pre-trained models become helpful, safe assistants.

Lesson 1 • Anatomy of an Effective Prompt

Breaks down the components of a well-formed prompt: instruction, context, input, and output format. Establishes a repeatable structure for all subsequent prompting work.

Lesson 2 • Zero-shot and Few-shot Prompting

Contrasts prompting with no examples versus providing demonstrations. Teaches when each approach maximises output quality.

Lesson 3 • Prompt Iteration and Debugging

Provides a systematic process for diagnosing and fixing underperforming prompts. Builds a habit of evidence-based prompt refinement.

Lesson 4 • Chain-of-Thought and Reasoning Prompts

Introduces step-by-step reasoning elicitation to improve accuracy on complex tasks. Connects reasoning prompts to measurable output quality gains.

Lesson 1 • Cost Monitoring and Optimisation

Teaches token-based pricing models and strategies to reduce API spend without sacrificing quality. Prepares learners for responsible production deployment.

Lesson 2 • Building Your First AI Script

Walks through a complete Python script that calls an LLM API and processes the response. Delivers a working artefact learners can extend immediately.

Lesson 3 • API Basics and Authentication

Covers REST API concepts, API key management, and rate limiting fundamentals. Establishes safe, functional access to AI services.

Lesson 4 • Core API Parameters Explained

Demystifies temperature, top-p, max tokens, and stop sequences with practical examples. Enables precise control over model output behaviour.

Lesson 1 • Document Ingestion and Chunking

Covers loading, cleaning, and splitting documents into retrieval-ready chunks. Quality at this stage directly determines retrieval accuracy downstream.

Lesson 2 • Evaluating and Improving RAG Pipelines

Introduces faithfulness, relevance, and answer correctness metrics for RAG evaluation. Provides a feedback loop for iterative pipeline improvement.

Lesson 3 • Retrieval Strategies and Reranking

Compares dense, sparse, and hybrid retrieval methods and introduces reranking. Teaches how to maximise the relevance of retrieved context.

Lesson 4 • Embeddings and Vector Stores

Explains how text is converted to vectors and stored for similarity search. Connects embedding quality to retrieval precision.

Lesson 1 • Tool Use and Function Calling

Covers how LLMs invoke external tools via structured function calls. Enables agents to interact with APIs, databases, and code executors.

Lesson 2 • Planning and Reasoning Frameworks

Introduces ReAct, plan-and-execute, and tree-of-thought agent architectures. Equips learners to select the right reasoning pattern for a given task.

Lesson 3 • Multi-Agent Orchestration

Covers orchestrator-worker patterns, agent communication protocols, and task delegation. Prepares learners to build collaborative agent networks.

Lesson 4 • What Makes a System Agentic

Defines agency in AI systems through planning, tool use, and feedback loops. Distinguishes single-turn LLM calls from true agentic architectures.

Lesson 1 • Parameter-Efficient Fine-tuning Methods

Introduces LoRA, QLoRA, and adapter methods that reduce compute requirements. Makes fine-tuning accessible without full-model training infrastructure.

Lesson 2 • When Fine-tuning Is the Right Choice

Establishes decision criteria for choosing fine-tuning over prompting or RAG. Prevents costly fine-tuning projects where simpler solutions suffice.

Lesson 3 • Running a Fine-tuning Job

Walks through configuring hyperparameters, launching training, and monitoring loss curves. Produces a reproducible fine-tuning workflow learners can reuse.

Lesson 4 • Evaluating and Deploying Custom Models

Covers task-specific evaluation metrics, regression testing, and serving fine-tuned models. Closes the loop from training to production deployment.

Lesson 1 • Safety, Guardrails, and Content Moderation

Implements input and output guardrails to prevent harmful, off-topic, or policy-violating content. Protects users and organisations from foreseeable AI risks.

Lesson 2 • Observability and Monitoring

Introduces logging, tracing, and alerting strategies specific to LLM-based systems. Enables rapid diagnosis of quality regressions and cost anomalies.

Lesson 3 • Data Privacy and Compliance Considerations

Addresses PII handling, data residency, and audit logging for regulated environments. Prepares learners to deploy AI within organisational compliance frameworks.

Lesson 4 • Continuous Improvement and MLOps

Establishes feedback loops, A/B testing, and retraining triggers for long-lived AI systems. Ensures deployed models improve rather than degrade over time.