Data Science interviews. 10 titles, each with its own model.
Each page shows the competency model, the questions that test it, what changes with seniority, and where candidates lose the interview.
Last reviewed
Every Data Science title
10 titlesWhat the report looks like for Data Science
- The situation in one sentence: where, what was short, and by how much.
- The decision that was yours rather than the team’s, and what you chose not to do.
- One number or one consequence that shows it worked.
Every competency is scored out of five with a line you actually said as evidence, and the weakest answers are rewritten as structure and specifics, never a script.
What Data Science interviewers keep scoring
Formulating the problem & choosing the approach
Converts a business objective into a well-defined analytical or predictive problem, chooses the simplest method that could work, and defines success in terms of the decision it supports.
Statistical inference & experimentation
Designs and analyses experiments and observational studies correctly: power, randomisation, multiple comparisons, confounding and causal inference where experiments are impossible.
Modelling, validation & error analysis
Builds predictive models with honest validation, understands their failure modes through error analysis, and prevents leakage and overfitting.
Data understanding & feature engineering
Investigates how data was generated before using it, engineers features grounded in domain knowledge, and writes reproducible data preparation code.
Taking models from prototype to production
Works with engineering to deploy and monitor models, writes code others can maintain, and understands the operational constraints of the systems that consume predictions.
Communicating insight & uncertainty
Explains findings, model behaviour and limitations to decision-makers so they act appropriately, and resists overstating certainty.
Ethics, privacy & fairness
Considers bias, fairness and lawful use of personal data (POPIA/GDPR) in analyses and models, and raises concerns even when inconvenient.
Scientific ownership & intellectual honesty
Owns the correctness of their work: seeks disconfirming evidence, admits when results do not hold, and follows a project through to real impact rather than a slide.
Problem framing & success metrics
Translates a business or product goal into a well-posed ML task with a metric that reflects real value, and knows when ML is not the right tool.
Data quality & feature engineering
Understands and validates the data before modelling, engineers features with awareness of leakage and drift, and reproduces data pipelines reliably.
Modelling & rigorous evaluation
Selects and trains appropriate models, evaluates them honestly with proper baselines and error analysis, and can explain why a model made a prediction.
Deploying & operating ML in production
Ships models as reliable services or batch jobs with monitoring for drift, latency and quality, and can retrain, roll back and explain incidents.