Interview prep
Data Scientist: 7-day sprint
Seven days on what data science loops test - statistics, A/B testing, metrics, classical ML, and product cases, plus enough LLM literacy for today's teams.
Rapid-fire this sprint0/29 done
Reading mode for lessons:
The job. Answer product questions with data: define metrics, run experiments, build models where they help, and explain the result to people who will act on it.
What the loop usually tests
| Round | What they probe | Where it's covered |
|---|---|---|
| Statistics | Distributions, hypothesis tests, confidence intervals, Bayes | Day 1 |
| Experimentation | A/B design, power, pitfalls, reading results | Day 1 |
| Modeling | Metrics, validation, classical models, interpretability | Days 2 to 4 |
| Product case | Metric definition, diagnosing a drop, deciding with data | Day 5 |
| LLM literacy | Evaluating and measuring LLM features | Day 6 |
Day 1: Statistics and experiments
- Read: Probability and statistics, Experimentation and A/B testing
- Lab: A/B test power and peeking
- Cheat sheet: formulas
- Be ready to: compute a sample size, explain a p-value without misstating it, and say why peeking inflates false positives.
Day 2: Validation and metrics
- Read: The language of ML, Workflow, splits, and leakage, Overfitting and bias-variance, Classification metrics, Regularization and regression metrics
- Also read: Imbalanced data
- Labs: underfitting vs overfitting, classification threshold
- Cheat sheets: classification metrics, regression metrics
- Be ready to: pick precision vs recall from business costs, spot leakage in a feature list, and choose MAE vs RMSE.
Day 3: The models you'll actually use
- Read: Linear regression, Logistic regression, Decision trees, Random forests and boosting
- Also read: Feature engineering, Hyperparameter tuning
- Labs: linear regression, decision-tree builder, gradient boosting
- Cheat sheet: model selection
- Be ready to: interpret coefficients (and say when you can't), explain feature importance pitfalls, and justify a GBDT.
Day 4: Segments, structure, and text
- Read: K-means, Density, hierarchical, and mixture clustering, PCA, Text to vectors
- Also read: Anomaly detection, Time series, Topic modeling
- Labs: K-means, PCA, TF-IDF
- Cheat sheet: clustering algorithms
- Be ready to: choose K and defend it, say why a discovered cluster isn't a real category, and build a quick text classifier baseline.
Day 5: Product and system cases
- Read: ML system design, Naive Bayes
- Also read: Monitoring and drift, Fairness
- Lab: recommendation funnel
- Cheat sheet: ML system design
- Be ready to: define a north-star metric with guardrails, and frame a churn or fraud problem as a model with a decision attached.
Day 6: Measuring LLM features
- Read: Adapting LLMs, The RAG pipeline, Evaluating LLM systems
- Also read: Prompt engineering
- Lab: RAG pipeline
- Cheat sheet: LLM evaluation
- Be ready to: design an offline eval for a summarization feature, validate an LLM judge against human labels, and plan the online test.
Day 7: Mock interview
- Rapid-fire: 20 mixed cards from this sprint, 60 seconds each (open the drill).
- Product case, 30 minutes: Weekly active users dropped 6% last week. Walk me through how you'd find out why.
- Experiment case, 30 minutes: We want to test a new checkout flow. Design the experiment, then interpret a result with p = 0.04 and a sample ratio mismatch.
- Modeling case, 30 minutes: Predict which subscribers will cancel next month, and tell me what the business should do with the scores.