The full curriculum

Start here: what you can skip

Get set up, and get honest about which of the next three modules you can skip.

AI in 2026: what actually changed - Tell the difference between what AI can genuinely do and what is marketing.
Set up Python and Google Colab - Run your first Python notebook in the browser, with no install. coming soon
Git and GitHub, the parts you use - Put code on GitHub so your work is visible and recoverable.
How to learn with AI without outsourcing your brain - Use an AI assistant so it accelerates you instead of replacing the learning.

Kill the vocabulary anxiety. No maths in this module.

AI vs ML vs deep learning vs LLM - Place the four terms correctly inside each other.
What machine learning actually is - Explain how a model learns a rule from examples instead of being given one.
Supervised vs unsupervised learning - Decide which of the two a given problem is.
Training vs inference - Say which costs money, which takes time, and which happens when you use ChatGPT.
Why AI took off now and not in 2005 - Name the three things that changed: data, compute, and the transformer.

Skip this if you already write Python. Do not, if you have shipped Java or Go for a decade.

Python crash course - Write and run Python confidently: variables, types, control flow.
Lists, dicts and loops - Reach for the right container and iterate over it cleanly.
Functions, imports and modules - Split code into functions and import across files.
Reading a Python traceback - Read an error bottom-up and fix it without pasting it anywhere.
Notebooks and virtual environments - Keep projects isolated and know when a notebook is the wrong tool.
NumPy arrays - Use arrays and vectorised operations instead of Python loops.
pandas DataFrames - Load a CSV, select rows and columns, and compute a grouped summary.

Enough intuition to read ML material. Explicitly not a maths degree.

Vectors, and what they actually represent - Read a vector as a point, a direction, and a list of features.
The dot product, and why similarity is just an angle - Explain why two similar things have a large dot product.
Matrices as transformations - See a matrix as something that moves space, not a grid of numbers.
Mean, variance and distributions - Describe a dataset's centre and spread, and read a histogram.
Probability, intuitively - Reason about likelihood and conditional probability without formulas.
What a derivative actually is - Read a derivative as a slope, which is all gradient descent needs.

From data to a trained model you evaluated honestly.

Features and labels - Turn a real problem into a table a model can learn from.
Loss: how wrong is the model? - Explain what a loss function measures and why training minimises it.
Gradient descent - Describe how a model walks downhill to better parameters.
Train/test split, and why it is non-negotiable - Evaluate a model on data it has never seen.
Overfitting and underfitting - Diagnose which of the two you have from train vs test performance.
Train your first model with scikit-learn - Train, predict and evaluate an actual model end to end.

Understand the thing everything else is built on.

What a neuron actually computes - Compute one neuron's output by hand: weights, sum, bias, activation.
Layers and activation functions - Say why a network without activations collapses to a single layer.
Backpropagation, intuitively - Explain how blame for an error is shared backwards through the network.
Build a neural network from scratch - Write a working network and its backward pass in plain Python.
PyTorch, first steps - Rebuild the same network in PyTorch and see what the framework does for you.

Open the box you have been using every day.

Tokens and tokenization - Explain why an LLM cannot reliably count the letters in a word.
Embeddings: meaning as coordinates - Describe how similar meanings end up close together in space.
Attention and the transformer - Explain what 'attention' lets a model do that earlier models could not.
Build GPT from scratch - Implement a small working GPT and understand every line.
Pretraining, fine-tuning and RLHF - Name the three stages that turn raw text prediction into an assistant.
Why LLMs hallucinate - Explain hallucination as a property of the objective, not a bug to be patched.

Go from typing into a chat box to calling a model from code.

What actually makes a prompt work - Write prompts with context, task and format instead of hoping.
System prompts, user prompts and roles - Put persistent instructions where they belong. coming soon
Getting structured output you can parse - Make a model return JSON your code can rely on.
Call an LLM API from Python - Make your first API call and handle the response in code.
Context windows, tokens and what things cost - Estimate the cost and the context limit of a call before you make it.
Write your first eval - Measure whether a prompt change actually helped.

Build the single most employable thing in this course.

Why RAG exists - Say what problem retrieval solves that a bigger model does not.
Chunking documents - Split a document so retrieved pieces are actually answerable.
Vector databases and similarity search - Store embeddings and retrieve the nearest ones to a query.
Build a document Q&A app - Ship a working app that answers questions about your own PDFs.
RAG vs fine-tuning: which one do you need? - Choose correctly between retrieval and fine-tuning for a given problem.

Let a model do things, and know how that fails.

What an AI agent actually is - Separate an agent from a chatbot and from a workflow.
Tool calling - Give a model a function and let it decide when to call it.
The agent loop - Trace think → act → observe → repeat, and where to stop it.
MCP: giving agents a standard set of tools - Explain what MCP standardises and why that matters.
How agents fail, and guardrails - Predict the common failure modes and bound them before shipping.

A URL someone else can open. This is the module that gives you something to point at.

Put a UI on it with Streamlit - Turn a script into an app someone can click.
Serve it as an API with FastAPI - Expose your work as an endpoint another program can call.
API keys, .env files and not leaking them - Keep credentials out of your repo, permanently.
Deploy it on a free tier - Get a public URL without paying for hosting.
Write a README that gets read - Make a stranger understand your project in thirty seconds.
Record a 60-second demo - Produce the artifact a hiring manager actually watches.

Convert what you built into leverage, where you already work or somewhere else.

Turn your work into a portfolio - Arrange your projects so a stranger sees competence in one screen. coming soon
Putting AI work on a senior resume - Describe a project in terms of the problem and the result.
Your first open-source contribution - Land a real pull request in someone else's project.
What interviewers actually ask - Answer the standard fundamentals confidently and out loud.
Limits, ethics, and what you must not put in a prompt - Know what you can and cannot send to a third-party model at work.
What to learn next - Pick your next track deliberately instead of drifting.