AI, demystified
Learn how AI actually works.
Twelve lessons on how these systems work, how you teach one your own material, how to tell whether it is any good, and how to run it without causing harm. Written for people who did not come from engineering, by a doctor who works on both sides of the clinical and AI divide.
No maths. No assumed background. Every technical term is defined the first time it appears, and you can hover any of them for a one-sentence reminder without losing your place.
Lessons 1 and 2 are free. Lessons 3 to 12 are part of the £30 playbook.
The short version first
Every section opens with a plain-English summary before any detail, so you always know what you are about to read. Switch it off once the vocabulary feels familiar.
Hover any jargon
Technical terms are underlined the first time they appear in a section. Hover or tap for a one-sentence definition without leaving the paragraph.
Check yourself as you go
Short questions at the end of each section, with the answers hidden. Trying to answer before you look is what makes it stick.
Nothing is dumbed down
Every concept from the technical version is still here, including the ones that sound intimidating. Only the sentences got simpler.
The 12 lessons.
Read in order, or jump aroundHow large language models actually work
What a model is actually doing when it answers you, from chunks of text and lists of numbers through to attention, training, and what a single answer costs.
Read free18 min readTeaching a model your stuff: prompting, RAG, and fine-tuning
The three ways to make a model useful for your problem: writing better instructions, giving it your documents to read, and retraining it. Plus how to tell which one you need.
Read free16 min readAgents, tools, and orchestration
What an AI agent actually is, how it takes actions, the patterns for coordinating several of them, and why acting on its own turns a small error rate into a big one.
Preview14 min readEvaluation: knowing whether it actually works
How to build a proper test set, the four numbers that diagnose a retrieval system, using a model to grade other models, and how to tell a real improvement from statistical noise.
Preview17 min readData: where the real advantage sits
Why the data layer is what actually defends an AI product: labelling, preference data, red-teaming, synthetic data, and deciding how much human oversight a task needs.
Preview12 min readSafety, guardrails, security, and sovereignty
Invented answers, prompt injection, the three places guardrails live, personal data and access control, where a system is allowed to run, and knowing when not to use AI.
Preview16 min readRunning AI in production without going broke
The quality, speed, and cost trade-off, working out what a system will actually cost, cutting waiting time, routing easy work to cheap models, and keeping the whole thing running.
Preview13 min readIs your AI actually being used, or just switched on?
The numbers that prove something is genuinely used rather than merely busy, how they are calculated, how to check a difference is real, and why the average is the most misleading figure on the dashboard.
Preview14 min readDesigning real systems: five worked architectures
Ten questions that design any AI system, applied to a tutor, a phone agent, a legal research assistant, a document pipeline, and a medical scribe.
Preview15 min readAnswering "why?" in two sentences
The recurring AI trade-offs, each with a short answer you can say out loud: retrieval or retraining, big or small, open or closed, agent or workflow, and more.
Preview9 min readThinking through real problems out loud
Worked answers to the questions that keep coming up: when to use retrieval, how to test a system, how to diagnose a broken one, when to refuse, and how to explain any of it to a non-technical audience.
Preview13 min readWhy AI in other languages is genuinely harder
Why non-English AI is harder: the token penalty, the gap between formal and spoken language, embeddings, scanning documents, and voice, plus how to design around all of it.
Preview11 min readFree vs full.
Unlock every AI lesson
One payment of £30 unlocks all 12 AI lessons in full, plus the entire Clinical Pivot Playbook: my story so far (later chapters are in draft), 8 career maps, 8 personalised 90-day plans, CV and LinkedIn templates, the cold-message bank, the interview answers, the glossary, and the startup list. No subscription.