teaching

Each course stands on its own and builds every idea by hand before handing it to a library. The notebooks open in Google Colab with one click, so there is nothing to install. The first two courses start from a first line of Python. The other three assume you can already program.

• From Python to Deep Learning and Fine-Tuning: for those heading towards machine learning and AI.
• From Zero to Data Structures and Algorithms in Python: a practice-first route through programming fundamentals, algorithms, and data structures.
• From Words to Meaning: Natural Language Processing in Python: how machines read, search, translate and summarise text, from counting words to retrieval-augmented generation.
• From One Table to a Cluster: Working with Data at Scale: SQL, data warehouses, NoSQL, Spark and streaming, and how to tell when you need them.
• From Random Moves to Winning Agents: Reinforcement Learning with Games: agents that learn by trial and error, from slot machines to Snake, a lunar lander and self-play, with videos of the agents you train.

Want to write your own code alongside the notebooks? Open a blank Colab notebook.

From Python to Deep Learning and Fine-Tuning

A complete, self-contained course that takes you from a first line of Python to fine-tuning a modern Transformer model. The notebooks run directly in the browser via Google Colab: nothing to install.

Each concept is first explained in depth, then implemented by hand, and only then handed to a library. That order is deliberate: it is what turns these systems from black boxes into things you genuinely understand.

Module A: Python Foundations

Module B: Scientific Python

Module C: Machine Learning

Module D: Deep Learning

Module E: Transformers and Fine-Tuning

From Zero to Data Structures and Algorithms in Python

A complete, practice-first course that takes you from a first line of Python to building your own stacks, linked lists, hash tables and search trees. It assumes no programming experience. The notebooks run directly in the browser via Google Colab: nothing to install.

Each concept is first explained in plain words, then shown in small worked examples, and then practised: the 18 notebooks contain more than 140 exercises, each with a hidden solution. Every claim about speed is measured rather than asserted, with timing experiments you run yourself.

Module A: Python from Zero

Module B: Mini Projects

Module C: Algorithms and Complexity

Module D: Data Structures

From Words to Meaning: Natural Language Processing in Python

A complete course on how machines work with human language, from a single string of characters to systems that search, translate, summarise and answer questions. It assumes you can already program in Python. The notebooks run directly in the browser via Google Colab: nothing to install.

One question runs through all 18 notebooks: how do we turn text into numbers without losing what it means? Each idea is first explained in plain words, then built by hand in a few lines of Python, and only then handed to a library. Every method is measured against a simple baseline, and the result is reported even when the simple method wins.

The course does not repeat the foundations of machine learning and neural networks. Those are covered in From Python to Deep Learning and Fine-Tuning, and the notebooks here point to the exact notebook there whenever they rely on it.

Module A: Text as Data

Module B: The Structure of Language

Module C: Counting Words

Module D: Meaning as Vectors

Module E: Neural Networks for Text

Module F: Language Models in Practice

From One Table to a Cluster: Working with Data at Scale

A complete course on what happens to data once there is a lot of it: SQL and indexes, data warehouses, columnar files, NoSQL, MapReduce, Spark, streaming and the cloud. It assumes you can already program in Python, and no knowledge of databases. The notebooks run directly in the browser via Google Colab: nothing to install.

One question runs through the first 12 notebooks: what changes when the data no longer fits? In one table, in the time a query has, in memory, on one machine, and finally in time itself. Each idea is first explained in plain words, then built by hand in a few lines of Python, and only then handed to a professional tool. Every claim about speed or size is measured, including the cases where the simple approach beats the heavy machinery. The six notebooks of the last module take one real data warehouse, Microsoft’s AdventureWorks sample, through every idea of the course.

The course does not repeat what the others cover. Working with tables in pandas and machine learning are in From Python to Deep Learning and Fine-Tuning. Big O, binary search, hash tables and trees are in From Zero to Data Structures and Algorithms in Python.

Module A: Data at Rest

Module B: Beyond Tables

Module C: Too Big for One Machine

Module D: Data in Motion and in the Cloud

Module E: A Real Warehouse, End to End

From Random Moves to Winning Agents: Reinforcement Learning with Games

A complete course on agents that learn by trial and error. Nobody shows them the right answer: they act, receive a reward, and work out for themselves how to get more of it. Every idea is taught with a game, from a row of slot machines to Snake, a lunar lander, Flappy Bird and Connect Four, and in almost every notebook you watch a video of the agent you have just trained. The notebooks run in Google Colab without a GPU.

Each method is first explained in plain words, then written by hand in a few lines of Python, and only then taken from a library: Q-learning, deep Q-networks, policy gradients, PPO, tree search and self-play. Results in this field vary a great deal from one run to the next, so the course measures before it claims, and says so when an experiment shows no clear difference.

The course assumes you can program in Python. From Module C on the agents are neural networks in PyTorch, which are covered in From Python to Deep Learning and Fine-Tuning. Most cells finish in seconds. A few train an agent for some minutes, and the notebook says so beforehand.

Module A: Foundations

Module B: Learning with a Table

Module C: Deep Value-Based Learning

Module D: Policy-Based Learning

Module E: Beyond a Single Agent