Citadel·Data Scientist·Onsite - System Design / Architecture·Senior

SeniorPrefer not to say

Jun 2026

Summary

Citadel data scientist interview with a single meaty system design question covering the full ML pipeline for real estate price prediction. The depth they expected was pretty intense, covering everything from feature engineering to fairness audits to deployment monitoring.

Questions Asked (1)

Design a complete machine learning system to predict residential property sale prices across a large city, covering feature selection and engineering, training and validation strategy, model choice, fairness considerations, and deployment and monitoring.

System DesignData ModelingTechnical Trade-offs

Author's notes

This was a beast of a question and I spent the first two minutes just trying to figure out where to start.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer as a coherent system design, starting with problem framing and data understanding, then moving through feature engineering, model training and validation, fairness, and finally deployment and monitoring. Emphasize trade-offs at each stage, such as model complexity vs. interpretability, and how you would validate and monitor the system in production.

Pro tip: At Citadel, they care about robustness and edge cases—mention how you'd handle data leakage, temporal validation, and distribution shifts, and propose a feedback loop for continuous improvement.

1. Problem Framing and Data Understanding

Define the prediction target (sale price), evaluation metrics (e.g., MAE, RMSE, MAPE), and business constraints. Identify data sources (property records, economic indicators, geospatial data) and assess data quality, volume, and potential biases.

2. Feature Engineering and Selection

Engineer features from raw data: property attributes (size, age, rooms), location (neighborhood, distance to amenities), temporal (market trends), and interaction terms. Use domain knowledge and techniques like target encoding for categorical variables, and select features via importance or regularization.

3. Model Training and Validation

Choose models (e.g., gradient boosting, regularized linear models) and set up a robust validation strategy: time-based split to mimic real-world forecasting, cross-validation with spatial grouping to avoid leakage. Tune hyperparameters and evaluate against baselines.

4. Fairness and Bias Mitigation

Assess fairness across protected groups (e.g., race, income level) by measuring disparities in prediction errors. Apply mitigation techniques like reweighting, adversarial debiasing, or post-processing, and document trade-offs between fairness and accuracy.

5. Deployment and Monitoring

Deploy as a batch or real-time API, with a pipeline for feature computation. Monitor performance (RMSE, drift), data quality, and fairness metrics. Set up alerts and a retraining schedule, and include a feedback loop for model updates.

Key Points to Mention

Handling data leakage and using temporal validation to simulate production
Feature engineering with geospatial and temporal data, including target encoding for high-cardinality categoricals
Model choice trade-offs: interpretability (linear models) vs. performance (gradient boosting, neural networks)
Fairness metrics and mitigation strategies, and the importance of documenting assumptions
Deployment architecture: batch vs. real-time, feature store, and API design
Monitoring for data drift, concept drift, and fairness, with automated retraining

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