Capital One·Machine Learning Engineer·Technical Phone Screen·Intermediate

Intermediate

May 2026

Summary

Capital One ML engineer interview with a practical ML design question. Nothing too wild but it required more depth than I expected going in.

Questions Asked (1)

You have a dataset of historical deliveries. How would you build a model to predict delivery times?

Data ModelingTechnical Trade-offsSystem Design

Author's notes

I went straight to regression and started listing features like distance, time of day, weather.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business objective and data characteristics, then outline a structured ML pipeline from data preparation to model deployment. Emphasize feature engineering, model selection with trade-offs, and evaluation metrics aligned with business impact.

Pro tip: Highlight the importance of temporal validation and avoiding data leakage, as delivery time prediction is inherently time-series. Also, discuss how you would handle real-world constraints like missing data and concept drift.

1. Understand the Problem and Data

Clarify the prediction goal (e.g., estimated time of arrival) and explore the dataset to identify relevant features, target variable, and potential issues like missing values or outliers.

2. Data Preparation and Feature Engineering

Clean the data, handle missing values, and create features such as distance, time of day, day of week, weather conditions, and historical averages. Consider encoding categorical variables and scaling numerical ones.

3. Model Selection and Training

Choose appropriate models (e.g., gradient boosting, random forest, or neural networks) and train them using a time-based split to avoid leakage. Perform hyperparameter tuning with cross-validation.

4. Evaluation and Validation

Evaluate models using metrics like MAE, RMSE, and MAPE, and validate on a holdout set that respects temporal order. Consider business-specific metrics like percentage of deliveries within a time window.

5. Deployment and Monitoring

Discuss how to deploy the model (e.g., API endpoint) and monitor its performance over time, including retraining strategies to handle concept drift.

Key Points to Mention

Temporal validation to prevent data leakage
Feature engineering with domain-specific variables (e.g., traffic, weather)
Handling missing data and outliers
Model choice trade-offs (interpretability vs. accuracy)
Evaluation metrics aligned with business goals
Deployment considerations and monitoring for drift

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.