Amazon·Data Scientist·Technical Phone Screen·Senior

Senior

Jul 2026

Summary

Amazon DS loop, and this question basically consumed the entire session. It was a deep optimization and ML design problem dressed up as an ops question, and I was not fully ready for how far they wanted to push it.

Questions Asked (1)

We used to assign packages to couriers manually. Design a system to do this automatically, optimizing for on-time delivery and fairness across couriers. Walk through your objective function, constraints, required data, algorithmic approach, and how you'd handle edge cases like new couriers, distribution shift, and last-minute volume spikes. Include pseudocode for the full pipeline and explain what happens if the optimizer fails.

System DesignTechnical Trade-offsAlgorithms & Data Structures

Author's notes

This was one question that just kept expanding.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business objective and defining a multi-objective function that balances on-time delivery and courier fairness, then outline the constraints, data requirements, and a two-stage algorithmic approach (assignment + routing). Walk through edge cases and failure modes, and provide pseudocode for the full pipeline, emphasizing monitoring and fallback mechanisms.

Pro tip: Frame fairness as a constraint or a secondary objective with a tunable weight, and discuss how you'd validate the system with A/B tests and counterfactual simulations before full rollout.

1. Clarify objectives and define the objective function

Ask clarifying questions about business goals (e.g., on-time rate target, fairness definition) and formulate a multi-objective function that combines on-time delivery probability and fairness metrics (e.g., variance in workload or earnings).

2. Identify constraints and required data

List hard constraints (courier capacity, vehicle type, time windows) and soft constraints (preferences, fairness bounds). Specify data needed: historical delivery times, courier availability, package attributes, traffic, weather, and real-time updates.

3. Propose algorithmic approach and pseudocode

Describe a two-stage approach: first, predict delivery times and on-time probabilities; second, solve an assignment problem (e.g., min-cost flow or Hungarian algorithm) with fairness constraints, followed by routing optimization. Provide pseudocode for the pipeline.

4. Address edge cases and failure modes

Explain how to handle new couriers (cold-start with exploration), distribution shift (online learning, periodic retraining), and volume spikes (dynamic re-optimization, overflow to manual). Detail fallback if optimizer fails: revert to manual assignment or a greedy heuristic.

5. Discuss evaluation, monitoring, and iteration

Outline offline evaluation (simulation, counterfactual), online A/B testing, and monitoring metrics (on-time rate, fairness, courier satisfaction). Emphasize continuous improvement and feedback loops.

Key Points to Mention

Multi-objective optimization: weighted sum or Pareto frontier for on-time delivery vs. fairness.
Fairness metrics: workload balance, earnings equity, or opportunity equality across couriers.
Two-stage approach: prediction (ETA, on-time probability) + assignment (min-cost flow, Hungarian) + routing (VRP).
Cold-start for new couriers: use exploration/exploitation, assign simpler routes initially, and learn quickly.
Distribution shift: monitor drift, retrain models frequently, use online learning for real-time adaptation.
Volume spikes: dynamic re-optimization, surge pricing, and fallback to manual or greedy assignment.

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