Mid-Level Data Scientist Interview Preparation Guide (FAANG Standard)

Be clear about your current employment status, notice period, and availability for interviews. Confirm you're available for the interview timeline and understand whether interviews will be virtual, hybrid, or onsite. Discuss any scheduling constraints upfront.

Prepare 2-3 concrete examples of projects where you had significant impact. For each, briefly describe the business problem, your role, the technical approach, and the results (e.g., 'Improved model accuracy by 15% leading to $2M annual revenue lift'). For mid-level, emphasize owning the full project scope, making trade-off decisions, and collaborating with other teams.

Articulate why you're interested in this specific role, team, and company. Connect your technical expertise to the problems the company solves. Mention specific aspects of the job description that excite you (e.g., building ML models for recommendations, analyzing user behavior at scale) and explain how they align with your career goals.

Clearly articulate your data science career journey, focusing on key milestones, growth areas, and progression from junior to mid-level responsibilities. Emphasize how you've transitioned from individual contributor work to owning full-cycle projects and mentoring junior team members. For mid-level, highlight your ability to design approaches, make independent decisions, and drive meaningful business impact.

Understand basic query optimization: using indices effectively, avoiding unnecessary joins or subqueries, filtering data early in the query, and understanding query execution plans. For mid-level, you don't need to be a database expert, but you should recognize inefficient patterns and suggest improvements if asked.

Practice problems simulating realistic business situations: combining partial address records from multiple sources, identifying a customer's first marketing touchpoint, calculating top earners by department, handling missing or malformed data, and dealing with time-zone differences or duplicate records. Always consider NULL values, empty result sets, and what happens when business assumptions don't hold.

Write queries that filter, aggregate, and group data efficiently. Use WHERE, GROUP BY, HAVING clauses correctly. Calculate metrics like counts, sums, averages, percentiles, and distinct values. Handle time-based filtering (e.g., 'data from last 30 days') and conditional logic (e.g., 'completed transactions only'). Practice combining multiple aggregations and filtering conditions in a single query.

Be comfortable with pandas DataFrames: filtering rows, selecting columns, groupby operations, merging/joining DataFrames, applying functions with apply() and map(), handling missing values, and reshaping data. Understand NumPy arrays and basic operations. Write readable code that could be understood by a colleague. Know when to use pandas vs. raw Python loops.

Understand and apply window functions like ROW_NUMBER(), DENSE_RANK(), RANK(), LAG(), LEAD(), and aggregate functions with OVER clauses. Practice partitioning by meaningful groups (e.g., department, user_id) and ordering by relevant criteria (e.g., salary, timestamp). Use window functions to find top N items per group, calculate running totals, or identify time-series patterns.

Master all JOIN types (INNER, LEFT, RIGHT, FULL OUTER) and understand when to use each. Know how to join multiple tables, handle NULL values, and deal with edge cases like duplicate keys or many-to-many relationships. Practice joining address tables by city IDs, linking users to sessions, and mapping transactions to customer attributes. Understand the difference between including or excluding unmatched records based on business requirements.

Understand confidence intervals: a 95% CI means if we repeated the experiment many times, 95% of intervals would contain the true parameter. Know the relationship between sample size, confidence level, and interval width. Discuss what 'statistically significant' means in context and why practical significance might differ.

Recognize that correlation (statistical relationship between variables) doesn't imply causation (direct cause-and-effect relationship). Identify confounding variables that might explain an observed correlation. For example, summer ice cream sales and crime rates both increase but neither causes the other—warm weather is the confounder. Practice identifying lurking variables in business scenarios.

Understand the CLT: when you take multiple random samples and calculate their means, those sample means are normally distributed (bell-shaped) even if the underlying data isn't. Know that normal distribution is parameterized by mean and standard deviation. Appreciate why this matters: it allows you to estimate population characteristics from samples and construct confidence intervals.

Understand the concept of hypothesis testing: formulating null and alternative hypotheses, calculating p-values, and interpreting results. Know that a p-value represents the probability of observing results at least as extreme as the data if the null hypothesis is true. For example, in an A/B test, if p-value = 0.04, there's a 4% chance the observed difference is due to random variation. Be clear on the threshold (typically 0.05) and what it means to reject or fail to reject the null hypothesis.

Clearly distinguish between Type I errors (false positives: rejecting a true null hypothesis) and Type II errors (false negatives: failing to reject a false null hypothesis). Understand the trade-off between alpha (acceptable false positive rate) and beta (acceptable false negative rate). Practice identifying these errors in business contexts (e.g., fraud detection: incorrectly flagging a legitimate transaction vs. missing actual fraud).

Design end-to-end A/B tests: define metrics and success criteria, calculate sample size based on baseline conversion rate and desired effect size, determine test duration, randomize users appropriately, and interpret results. Discuss potential pitfalls: peeking at results early, running too short, insufficient sample size, and seasonal confounds. Be prepared to recommend whether to ship a feature based on test results.

Understand RNNs process sequential data by maintaining hidden state across time steps. Know they're used for language translation, voice recognition, and time series prediction. Appreciate limitations (vanishing gradients) and variants (LSTMs, GRUs) that address them. For mid-level, knowing when sequential models apply is more important than implementing them.

Use cross-validation (k-fold, stratified, time series) to estimate model performance fairly. Understand train/validation/test splits and why each is important. For time series, use forward chaining (don't peek into future). Discuss why simple train/test split can be misleading and when cross-validation is essential.

Understand neural network basics: layers (input, hidden, output), activation functions (ReLU, sigmoid), forward pass, backpropagation. Know that deep learning excels at learning hierarchical representations from raw data (images, text). Discuss when deep learning is justified (large datasets, complex patterns) vs. overkill (small datasets, interpretability required). Awareness of common architectures (CNNs for images, RNNs for sequences) is important.

Understand when to use different algorithms: logistic regression (interpretable, fast), decision trees (intuitive, prone to overfitting), random forests (robust, less interpretable), neural networks (complex patterns, data-hungry), SVMs (high-dimensional data), KNN (simple baseline), gradient boosting (often best in competitions). Know trade-offs: accuracy vs. interpretability, training time vs. performance, data requirements vs. model capacity. For mid-level, you should recommend appropriate algorithms for given problems and explain your reasoning.

Select appropriate metrics for the problem: accuracy (classification), precision/recall (imbalanced classes), F1 score (balance false positives/negatives), AUC (ranking models), RMSE (regression), MAPE (time series). Understand when each metric matters—precision is crucial for fraud (false alarms are costly) while recall matters for disease detection (missing cases is worse). Use multiple metrics to evaluate model; a single metric can hide problems. Discuss baseline and champion models.

Understand overfitting (model fits training data too closely, poor generalization) and underfitting (model too simple to capture patterns). Discuss regularization techniques: L1/L2 regularization (penalize complex models), dropout (neural networks), early stopping (boosting), and simpler models. Recognize signs of overfitting (high training accuracy, low test accuracy) and strategies to combat it.

Feature engineering is often the most impactful ML task. Discuss creating features from raw data: numerical transformations (scaling, binning, logarithms), categorical handling (one-hot encoding, target encoding), temporal features (time of day, day of week, seasonality), and domain-specific features. Understand that good features should be predictive, not redundant, and interpretable. Create feature vectors (n-dimensional numerical representations) that feed into ML models. For mid-level, show you can think creatively about features that might drive predictions.

Recognize common data quality issues: missing values, duplicates, malformed data, schema changes, delays in data pipelines. Discuss validation: checking record counts, verifying data distributions, confirming joins are correct, testing for unexpected nulls. When metrics look wrong, first suspect data quality before blaming the product. Develop debugging habits: compare to historical baselines, segment data, and check upstream data sources.

Understand how to track where customers come from. In attribution, you want to assign credit to touchpoints that led to conversion. For example, a customer might visit through organic search, then through a paid ad, then directly—which channel deserves credit? Discuss attribution models: first-touch (credit first channel), last-touch (credit final channel), linear (distribute equally), time-decay (credit more recent interactions). Understand trade-offs and limitations.

Communicate findings to non-technical audiences: product managers, executives, marketers. Translate technical results into business language. Lead with the insight (e.g., 'New onboarding flow reduced churn by 8%'), then explain the supporting data, methodology, and caveats. Use visualizations effectively—graphs often communicate faster than tables. Discuss limitations and uncertainty honestly. Be prepared to defend recommendations against skepticism.

Approach ambiguous problems systematically. Start by clarifying the problem statement: Is it a real problem? How big is it? Is it sudden or gradual? Then decompose: break the metric into sub-components, segment users (new vs. existing, geographic, device), compare to baselines (day-over-day, week-over-week, year-over-year). Use a framework: macro (market, seasonality), micro (product, feature), technical (systems, infrastructure). Prioritize which hypotheses to investigate based on impact and likelihood.

Clearly define business metrics: what you're measuring and why. Common examples include conversion rate, churn rate, daily active users (DAU), lifetime value (LTV), retention, engagement, click-through rate (CTR). Understand the difference between metric (what you measure) and target (what you aim for). Discuss leading indicators (predict future outcomes) vs. lagging indicators (measure past performance). Know how to decompose complex metrics (e.g., revenue = DAU × engagement × monetization).

Design experiments to answer business questions: 'Should we ship this feature?' 'Which design performs better?' 'Does this pricing change increase revenue?' Define success metrics, estimate effect size, calculate sample size and test duration, randomize properly, monitor results, and make decisions. Discuss potential pitfalls: peeking bias (looking early), network effects (can't isolate users), time-horizon effects (short-term vs. long-term impact).

Show you understand and align with the company's culture and values (e.g., FAANG companies often value user-centricity, bias for action, data-driven thinking). Reference company examples or products in your answers. Ask thoughtful questions about how the team embodies company values. Demonstrate genuine interest in the company beyond just the job.

Discuss a project or analysis that didn't go as planned. Explain what happened, what you learned, and how you applied that learning. For mid-level, show maturity: acknowledge the setback, focus on learning over blame, and explain concrete changes in your approach. Resilience and growth mindset are important.

Describe situations where direction was unclear or you disagreed with a colleague's approach. Show how you gathered information, advocated respectfully for your viewpoint, and reached a decision. For mid-level, demonstrate that you can disagree constructively, consider other perspectives, and move forward even without total alignment. Don't just be agreeable; show healthy disagreement.

Discuss experiences mentoring junior colleagues, helping teammates upskill, or sharing knowledge. For mid-level, this might be informal (helping a junior colleague debug code, teaching a technique) or more structured (onboarding a new team member, leading a knowledge-sharing session). Show you're invested in others' growth and can explain complex concepts clearly.

Share examples of collaborating with engineers, product managers, business teams, or other functions. Describe how you communicated complex ideas to non-technical audiences, adapted your communication style, and worked toward shared goals. For mid-level, emphasize how your insights influenced decisions or how you navigated conflicting priorities. Show you're collaborative, not siloed.

Demonstrate how you've owned projects from ideation through delivery and measurement. For mid-level, showcase a project where you drove the decision (not just executed), navigated ambiguity, overcame obstacles, and measured impact. Include what went well, what you'd do differently, and lessons learned. Quantify outcomes: 'Improved model accuracy by 15%, leading to $2M revenue lift.' Show accountability and ownership mentality.