About Data Academy

Data Academy is an independent publication about data science as it’s actually practised — not as it’s marketed. The gap between the two is where most newcomers get stuck, and where plenty of experienced people quietly waste years.

The emphasis here is deliberately unfashionable: problem framing, data cleaning, validation, evaluation, SQL, simple models, and clear communication. These are the skills that decide whether a data project succeeds, and they’re exactly the ones flashy tutorials breeze past on the way to the deep learning demo.

Alongside that sits the Foundations series: 33 first-principles explanations of the core ideas in data science and AI, from Bayes' theorem and the central limit theorem through to neural networks, transformers and diffusion models. Each one starts from scratch and stops before the hand-waving. Knowing why a method works is what stops it being used where it doesn’t.

Despite the name, this is not a course. There are no modules, no certificates, and nothing to enrol in — just articles, written the way a senior colleague would explain things: concrete, jargon-light, and biased towards what works in production.

Everything here is free. No sign-up, no newsletter, no tracking beyond what a static host records.

How these articles are written

The publication is anonymous on purpose, so it cannot lean on anyone’s reputation. What replaces a byline is method, and it is worth stating plainly:

• Code is run before it ships. Snippets are executed, not sketched. Numbers quoted from a worked example come from actually running it. Where a figure is illustrative rather than measured, it says so.
• Claims are sourced. When an article explains an idea that came from a specific piece of work, it names that work — Shannon on entropy, Breiman on random forests, Bickel on the Berkeley admissions data. API behaviour links to the official documentation rather than to a summary of it.
• Uncertainty is not smoothed over. Where practitioners genuinely disagree, or where a method's assumptions are doing more work than its maths, the article says that instead of picking the tidier story.
• Nothing is written to fill a schedule. There is no publishing cadence to hit, so there is no reason to pad. An article exists because there was something specific to say.

Corrections

Some of this will be wrong. When an error is found in something substantive, the article is corrected, it carries an updated date under its title, and the change is written down on the corrections page with what it used to say — so a page that has changed says so rather than pretending it always read that way.