- Dynamic Programming – Breaking down complex recursive problems into highly optimized iterative solutions using memoization or tabular methods.
- Python Internals – Understanding how built-in data structures (like lists and dicts) perform under the hood and knowing when to use specialized libraries.
Example scenarios:
- "Given a list of transaction dependencies, determine the correct execution order and identify if any circular dependencies exist."
- "Optimize a memory-intensive data processing script to run within a highly constrained container environment."
SQL & Product Metrics
As a Data Scientist, you must be exceptionally proficient in extracting insights from raw database tables. The SQL round evaluates your ability to write complex queries quickly and accurately, as well as your capacity to design product metrics from scratch.
The interview typically begins with standard querying tasks involving multiple table joins, aggregations, and window functions. Once you demonstrate technical proficiency, the round transitions into an open-ended case study where you must define key performance indicators (KPIs) for a specific product feature and then write the exact SQL queries required to calculate those metrics.
Be ready to go over:
- Advanced Aggregations & Window Functions – Using functions like
LEAD, LAG, RANK, and rolling cumulative sums to analyze sequential user behavior.
- Metric Framework Design – Translating vague business goals (e.g., "improve user engagement") into concrete, measurable metrics.
- Query Optimization – Writing efficient SQL that minimizes execution time and resource consumption on massive datasets.
Example scenarios:
- "Write a query to identify users who have experienced a sudden drop in transaction frequency compared to their personal 30-day average."
- "Define the metrics needed to evaluate the success of a new crypto trading feature and write the query to extract them."
Machine Learning Fundamentals & Math
This round is highly theoretical and separates average candidates from exceptional ones. Revolut expects its data scientists to understand the deep mathematical mechanics of the models they deploy.
You will face rapid-fire questions covering probability theory, statistics, and supervised/unsupervised learning algorithms. You must be prepared to write out mathematical proofs, derive loss functions, and explain optimization techniques on a virtual whiteboard.
Be ready to go over:
- Mathematical Derivations – Deriving formulas for standard algorithms, such as logistic regression, SVMs, or cross-entropy loss.
- Probability & Statistics – Applying Bayes' theorem, understanding probability distributions, and explaining hypothesis testing frameworks.
- Model Evaluation Mechanics – Explaining the exact mathematical differences between evaluation metrics (e.g., Precision-Recall AUC vs. ROC AUC) and when to use each.
Example scenarios:
- "Derive the gradient update rule for a linear regression model using mean squared error loss."
- "Explain how you would mathematically handle extreme class imbalance when training a neural network for fraud detection."
Machine Learning System Design
The system design round evaluates your ability to architect scalable, robust, and production-ready machine learning pipelines. You must demonstrate that you can think like an engineer as well as a scientist.
You will be given an ambiguous, high-level business problem and asked to design an end-to-end solution. This includes data ingestion, feature engineering, model selection, deployment strategies, real-time serving infrastructure, and post-deployment monitoring.
Be ready to go over:
- Feature Store & Pipeline Architecture – Designing scalable pipelines for both real-time feature extraction and batch processing.
- Model Deployment & Serving – Choosing between batch scoring and real-time inference based on latency and infrastructure constraints.
- Monitoring & Maintenance – Setting up automated systems to detect feature drift, concept drift, and model performance degradation in production.
Advanced concepts (less common):
- Designing distributed training setups for massive deep learning models.
- Implementing privacy-preserving machine learning techniques for sensitive financial data.
Example scenarios:
- "Design an end-to-end system to dynamically price international money transfers based on market volatility and user behavior."
- "Architect a real-time recommendation engine for the Revolut app homepage that scales to millions of active users daily."
