Data Scientist Interview Guide

Python fluency with relevant libraries: scikit-learn, PyTorch, TensorFlow, pandas, NumPy

Statistical foundations: regression, classification, clustering, hypothesis testing

Experience deploying models to production — not just Jupyter notebooks

Business impact framing: what business decision did your model inform or improve

SQL proficiency — nearly every data role requires it

For manufacturing/semiconductor: domain knowledge in process optimisation or yield improvement

Business impact — have you shipped models that improved a measurable business outcome?

Production ML experience — notebooks are not models; can you deploy, monitor, and retrain in production?

Statistical rigour — do you know when a result is statistically meaningful and when it is noise?

Domain expertise in the company's sector — a data scientist who understands yield loss is more valuable than a generalist

Communication — can you explain model results to a non-technical stakeholder without losing accuracy?

Statistical analysis and hypothesis testing methods you have applied

ML algorithms by type: regression, classification, clustering — be specific

Data wrangling and exploratory data analysis (EDA)

Big data tools: Spark, Hadoop, or similar at scale

Cloud platforms: AWS, GCP, or Azure at a practitioner level

Version control (Git) and reproducibility practices

Data ethics and privacy awareness — especially relevant under EU law

CVs that are all research with no deployed impact

Omitting the scale of data worked with: rows, frequency, pipeline complexity

Listing accuracy metrics without explaining the baseline or business value

No mention of how models were monitored or maintained after deployment

Q1

Walk me through a machine learning project you owned end-to-end, from problem definition to production.

Q2

How do you handle class imbalance in a classification problem? What metrics do you use to evaluate the model?

Q3

Explain how you would detect and handle data drift in a production ML model.

Q4

You have a feature with 40% missing values. Walk me through your decision process for handling it.

Q5

Tell me about a time your model performed well in development but poorly in production. What happened?

Structure each project as: problem → approach → outcome (with numbers)

Mention any MLOps experience: MLflow, Kubeflow, Weights & Biases

A Kaggle ranking or competition participation is a recognized signal in the Netherlands

For ASML or manufacturing roles, emphasise sensor data, time series, and anomaly detection experience

Customer churn was rising and the business had no reliable way to identify which customers were at risk.

The data was messy, incomplete, and spread across multiple systems with no clear churn signal.

Built a churn prediction model using classification algorithms, engineered features from raw event data, and created a monitoring dashboard for the commercial team.

Reduced churn by 12% within two quarters. Model deployed to production and still running.