New Course
From prompting to production
This course bridges the gap between casual AI use and building real systems. Covers prompt engineering, RAG, agents, MCP, evals, and multi-agent workflows.
What You'll Learn
Core modules
Eight modules covering the full stack of working with LLMs — from writing better prompts to orchestrating multi-agent systems.
Prompt Engineering
Design, template, chain, and evaluate prompts systematically.
RAG
Retrieval-augmented generation — from concept to optimisation.
Agentic Workflows
Multi-step autonomous systems, tool use, APIs, and MCP.
Evals & Multi-Agent
Measure what works, debug what doesn't, and scale with multiple agents.
Course Includes
Concept boardsCore ideas explained visually
FlashcardsKey terms to review per lesson
Practice questionsCheck understanding as you go
Video resourcesEmbedded YouTube per lesson
Additional readingPapers and articles per topic
Timed final exam20 questions, 30 minutes
Printable certificateUnlocked by passing the exam
Ready To Start
Begin with Lesson 1
Start with why base LLMs fall short, then work through prompt engineering, RAG, agents, and beyond.
Prerequisites
Who this course is for
This course assumes you have used an AI chatbot at least a few times. If you are a complete beginner, start with AI for Absolute Beginners first. If you mainly want to use agents safely rather than understand how they are built, AI Agents for Everyday Use is the shorter, practical route. No coding experience is required — the course explains technical concepts in plain language with practical examples.
Privacy & Storage
Course progress, exam results, and certificate details are stored in this browser on this device. They may be removed if you clear browser data.
All Lessons
Course Map
0 of 17 lessons complete
ExamNot attempted
Resume atLesson 1
CertificateLocked
Module 1 — The Problem10 min
Lesson 1: Why Base LLMs Aren't Enough
Pre-trained models are impressive, but they ship with real limitations that matter the moment you try to use them for serious work.
Module 1 — The Problem10 min
Lesson 2: Context Windows and Attention
Even the best models can only hold so much text in working memory, and they do not pay equal attention to all of it.
Module 2 — Prompt Engineering10 min
Lesson 3: Basic Prompt Design
The difference between a weak prompt and a strong one is not talent — it is structure. Role, task, context, and format do most of the work.
Module 2 — Prompt Engineering10 min
Lesson 4: Prompt Templates and Few-Shot Prompting
Templates let you scale a good prompt across thousands of users. Few-shot examples let you align the model to your definitions without touching its weights.
Module 2 — Prompt Engineering10 min
Lesson 5: Chain of Thought and Prompt Chaining
Chain of thought makes the model reason step by step. Prompt chaining splits a complex job into separate prompts you can debug independently.
Module 2 — Prompt Engineering10 min
Lesson 6: Testing and Evaluating Prompts
A prompt is only as good as the evidence that it works. Manual review gets you started; LLM judges let you scale.
Module 3 — Fine-Tuning9 min
Lesson 7: Fine-Tuning — Mostly Don't
Fine-tuning modifies the model's weights to specialise it for your task. It sounds powerful, but for most teams, the costs outweigh the benefits.
Module 4 — RAG10 min
Lesson 8: What RAG Is and Why It Exists
Retrieval-augmented generation connects an LLM to external knowledge so it can give grounded, sourced, up-to-date answers.
Module 4 — RAG10 min
Lesson 9: How RAG Actually Works
Understanding the mechanics — embeddings, vector databases, retrieval, and prompt templates — turns RAG from a buzzword into something you can reason about.
Module 4 — RAG10 min
Lesson 10: Improving RAG
Vanilla RAG is a starting point. Chunking, hierarchical retrieval, and techniques like HyDE can dramatically improve accuracy.
Module 5 — Agentic AI Workflows10 min
Lesson 11: From One-Shot to Multi-Step
An agentic workflow chains multiple steps — retrieval, tool use, follow-up questions, and actions — into a system that can handle real tasks end to end.
Module 5 — Agentic AI Workflows10 min
Lesson 12: From Fixed Rules to Fuzzy Systems
Building with AI means thinking differently about software — from deterministic systems where every step is predictable to fuzzy systems where outputs vary and guardrails matter.
Module 5 — Agentic AI Workflows10 min
Lesson 13: Agent Components
An agent is built from three core parts: memory (what it knows about you), tools (what it can do), and prompts (how it reasons). The balance between them determines how autonomous it is.
Module 5 — Agentic AI Workflows10 min
Lesson 14: APIs vs MCP
Traditional API integration is one-off and hard-coded. The Model Context Protocol (MCP) lets agents negotiate with tools dynamically, which scales much better.
Module 6 — Evals12 min
Lesson 15: Evaluating Agentic Workflows
Knowing whether your AI system works requires measuring it from multiple angles — end-to-end results, individual components, objective checks, and subjective judgement.
Module 7 — Multi-Agent Systems10 min
Lesson 16: Multi-Agent Systems
When a single agent cannot handle everything, you split the work across specialised agents that communicate and coordinate — just like a team of people.
Module 8 — What's Next10 min
Lesson 17: The Frontier
AI is moving fast. Scaling laws, architecture search, multimodality, and new learning methods will shape the next wave — and much of what we learned today will still apply.