Data Scientist interview questions

I've been a data scientist for six years, two in Oman. Started in analytics at an Indian fintech, transitioned to data science as ML became central, and for the past two years I've been data scientist at an Omani financial institution. My work: credit risk modelling, customer segmentation, fraud detection, churn prediction. Stack: Python (pandas, scikit-learn, XGBoost), SQL, MLflow for experiment tracking, model deployment via APIs. Master's in statistics. Continuing learning on deep learning and MLOps.

Problem framing first: classification, regression, ranking, etc. Data exploration: distributions, missing values, outliers, target balance. Feature engineering grounded in business understanding. Train/validation/test splits with appropriate methodology (temporal for time-series, stratified for imbalanced). Model selection from simple to complex; only escalate if simpler doesn't satisfy. Hyperparameter tuning with proper CV. Evaluation aligned with business metric. Documentation throughout. Modelling is craft; rushing produces models that fail in production.

Models deployed as APIs (FastAPI common). Containerised for portability. CI/CD pipeline includes model validation: performance threshold, data schema check. Monitoring in production: prediction distribution, data drift, performance degradation. Retraining triggered by drift or scheduled. Rollback path. Model deployment is engineering work; data science teams that don't think this way produce models that never reach production.

Interpretability matters more when stakes are high (credit decisions, regulatory contexts). SHAP values for tree models, surrogate models for complex ones, native interpretability for linear models. Business stakeholders need to understand model behaviour, especially for adverse decisions. Regulatory contexts often require interpretability for explanation requirements. Black-box models in regulated contexts create risk.

Translate between business question and ML approach. Business stakeholders care about decisions and outcomes; the model details matter less than the trade-offs. I prepare clear summaries: what the model predicts, what business action it enables, what the limitations are. I'm patient with non-technical questions. The relationship matters; stakeholders who trust models use them.

For a senior data scientist role at an Omani financial institution I'd target OMR 1,800 to 2,400 total package depending on the team's model portfolio and business impact. Roles with significant production-ML responsibility pay more. I'd expect annual bonus and continued professional development. I'm on 30-60 days' notice. Beyond pay I'd value the team's ML maturity; teams that ship models to production produce different careers than teams that just do notebook analysis.