Professional Certificate in Data Science

You need to have run a piece of Python once. That is the entire bar, and Module 0 Lesson 3 gets you there from nothing installed in about four minutes. You will not be writing Python from scratch at any point in this programme — an AI assistant writes it and you read it, interrogate it, and decide whether to trust it. That is not a beginner's compromise; it is how a great many working data scientists operate now, and the reading skill is the one that actually protects the work. Module 1 Lesson 3 is dedicated to it: a five-question read-through you run on any script an assistant hands you, including the technique of making the code print its own arithmetic so a silently dropped row cannot hide.

One verb. Data Analysis ends at interpreting existing data and models — framing questions, cleaning, descriptive statistics, reading a regression, reading an A/B test, SQL, dashboards, executive narrative. This certificate begins at constructing a model that scores cases nobody has observed yet. The Data Analysis programme's own Module 5 draws that line explicitly and hands you over. Where the two touch, this programme bridges by name rather than re-teaching: you already know how to frame a question, so we go straight to what changes when the answer has to be a model — the target definition, the prediction window, the baseline, and whether each column was even knowable at the moment you score. If you have not done the Data Analysis certificate you have lost nothing; Module 0's self-assessment tells you honestly where you stand.

About four weeks at a steady pace — roughly 26 hours split between the lessons and the hands-on work, across 47 lessons. It is fully self-paced and built to be stopped and restarted: cumulative reviews sit at the one-third and two-thirds marks specifically because motivation dips there in any multi-week programme, and each one consolidates what you have built before the next stretch. The capstone rewards learners who take their time over it rather than sprinting for the credential.

The evidence points somewhere more interesting than yes or no. Measured on realistic data-analysis tasks, capable models still solve only a minority of problems end to end, and their signature failure is code that runs perfectly while answering the wrong question. In one randomised trial, experienced developers using AI tools were about 19% slower — while believing they had been roughly 20% faster. Studies of people reviewing AI-generated analyses find professionals routinely miss decision-critical flaws, including ones stated in plain language rather than buried in code. What that adds up to is not replacement but a shift in where the value sits: from producing the analysis to verifying it. This programme is built entirely around that shift, which is why every module teaches the check alongside the technique.

No. Every exercise across all 47 lessons runs on free tiers: a free Claude, ChatGPT or Gemini account for the assistant, Google Colab for the notebook (free with any Google account, with pandas, numpy and scikit-learn pre-installed), and Streamlit Community Cloud for the deployment module. There is nothing to install and no credit card anywhere. Paid AutoML platforms are covered in Module 5 so you can evaluate them at work and understand what they automate and what they hide, but they are taught as things to assess rather than things to buy.

No, and the omission is deliberate rather than a gap. You will not derive an estimator, prove a convergence result, or meet measure-theoretic probability. What you will do is everything a working practitioner does around the algorithm: define the target, audit for leakage, choose the validation scheme, pick the metric from the decision, calibrate the probabilities, place the threshold where the costs say, read the errors, and document the whole thing. That is where applied projects actually succeed or fail. The field mastery map behind this programme tags the deep theory explicitly as out of scope with the reason attached — the craft is practised without it, and requiring it would gate out the readers this programme is built for.

Artefacts, not a badge. A written project brief with a target defined as a rule and a baseline stated before any code. A leakage audit you ran yourself. A reproducible feature pipeline. A tuned model with an honest held-out score and a confidence interval on it. An experiment design with a computed sample size and a pre-registered stopping rule. A cost-justified decision threshold. A per-slice performance table including the slice where your model is bad. A deployed app with a URL. A drift check. A completed model card. A written deployment recommendation. And a capstone you scoped and solved alone on a business you had never seen.

It carries a verifiable credential ID for your professional profile, but the honest answer is that the artefacts do more work than the certificate does. In a hiring conversation, walking someone through the leakage you found in your own data, the baseline you measured before you built anything, and the slice where your model underperforms is a far stronger signal than a line on a CV — because those are precisely the things that separate someone who has shipped a model from someone who has completed a tutorial. It is not an academic qualification and does not replace one. What it replaces is the portfolio gap.

A general assistant — Claude, ChatGPT or Gemini — for every module, deliberately tool-agnostic because the discipline transfers and the tools will not be the same in three years. Inside Colab you will also meet Google's built-in Gemini data science agent, which will plan and write an entire analysis notebook from a described goal; it is genuinely capable, and Google's own material notes plainly that it may make mistakes, which is the best possible argument for the reading skill this programme teaches. Module 5 additionally surveys the AutoML landscape, from free open-source options through to enterprise platforms, so you can judge them rather than be sold them.

It is written for you specifically, and the whole design follows from that. The reader this programme assumes can read a chart, reason about a business question, read SQL, and interpret a regression someone else fitted — and has written perhaps twenty lines of Python in their life. Everything follows from that: AI writes the code, the load-bearing skill is reading it critically, every term of art gets defined on first use, and the domain jargon is treated as the bigger wall than the AI jargon because for career-changers it usually is. What the programme does not do is pretend the modelling decisions are easy. They are the hard part, they are where projects die, and they are the part an analyst is unusually well-equipped to learn.

Not as buildable techniques, and Module 5 explains why in some detail. For the tabular business data this programme is about — rows and columns out of operational systems — gradient-boosted trees remain the working default, and reaching for a neural network usually buys nothing while costing a great deal. So deep learning is named, placed accurately on the map, and handed to the Master Certification, where it belongs alongside production machine learning and causal inference at depth. Being clear about where a technique stops being the right answer is part of the craft rather than an omission from it.

A Master Certification that picks up exactly where this one stops: causal inference beyond difference-in-differences, variance reduction and sequential testing, deep learning and foundation models, production machine learning as an engineering discipline, and leading the work rather than doing it. That is a genuinely different craft and an optional one — you do not need it to be excellent at this job. The more important thing this programme leaves you with is in Module 0 Lesson 4: the method for having an AI teach you a technique nobody wrote a lesson about, and verifying what it tells you in a subject you do not yet know. The capstone's final lesson marks exactly where you stand on the map and what the next steps are.