Data Science in Practice

What you will be able to do

• You will be able to design a trustworthy A/B test end to end, choosing the randomization unit, a primary metric or overall evaluation criterion, guardrail metrics, and a sample size backed by a real power calculation.
• You will be able to analyze experiment results honestly, separating statistical from practical significance, avoiding traps like peeking and multiple comparisons, and choosing a quasi-experimental design when randomization is not possible.
• You will be able to communicate findings to non-technical stakeholders with a bottom-line-first structure that ties every result to a decision and a recommended action.
• You will be able to visualize data and uncertainty accurately, picking the right chart type and showing confidence intervals without misleading the reader.
• You will be able to deploy a model into production with an appropriate serving pattern (batch, online, or streaming) and prevent training-serving skew using a feature store and model registry.
• You will be able to monitor a live model for data drift, concept drift, and performance decay, and define concrete retraining and rollback triggers.
• You will be able to frame a data science problem around business value, weighing expected return against cost and recognizing when machine learning is the wrong tool.
• You will be able to reduce responsible-practice risk by applying fairness metrics and their trade-offs, privacy obligations such as GDPR, and transparency artifacts like model cards and the EU AI Act.

What is inside

4 modules, 8 lessons. Each module ends in a graded quiz and most carry an assignment.

1. 01

   Designing Experiments and A/B Tests

   Most decisions that claim to be data-driven rest on experiments that quietly break their own assumptions. This module teaches you to design an A/B test you can defend (a falsifiable hypothesis, the right randomization unit, an overall evaluation criterion, guardrail metrics, and a sample size backed by a real power calculation) and then to read the results honestly: what a p-value does and does not say, why peeking and multiplicity manufacture false wins, and which quasi-experiment to reach for when clean randomization is off the table.

   2 lessons · 5 quiz questions

2. 02

   Communicating Results to Stakeholders

   A finding that no one understands or trusts changes nothing. This module turns analysis into a narrative that busy, non-technical stakeholders can act on: lead with the bottom line, tie every result to a decision, translate findings into business units, and visualize data and uncertainty so your charts inform the reader rather than quietly mislead.

   2 lessons · 5 quiz questions · assignment

3. 03

   Deploying and Monitoring Models

   The path from a working notebook model to a reliable production service, and how to keep it healthy once it is live. This module covers choosing a batch, online, or streaming serving pattern; packaging a model behind an API inside a container; using feature stores and a model registry to prevent training-serving skew; and monitoring for data and concept drift with clear retraining and rollback triggers.

   2 lessons · 5 quiz questions

4. 04

   Ethics and the Business Context

   Senior data scientists are judged on judgment, not just models. This capstone connects your work to business value: framing problems around the decisions they change, weighing return against total cost, and recognizing when machine learning is the wrong tool. It then turns to responsible practice: fairness metrics and their trade-offs, GDPR privacy obligations, and transparency tools including model cards and the EU AI Act.

   2 lessons · 5 quiz questions · assignment