Design Last Mile Delivery

Problem Statement: Last-Mile Parcel Delivery Platform

Last-mile connects sort centers, delivery associates, and recipients for the final leg after line-haul. Unlike ride-hailing, you optimize stop sequences, time windows, and proof-of-delivery under hard SLA tiers (same-day, next-day, scheduled).

Interviewers expect Shipment Command as the single writer; dispatch routes and ETAs are eventually consistent projections keyed by shipment_id.

How the pieces fit

A parcel flows from line-haul into a sort center — scanned, sorted to a route, and loaded — then a delivery associate pulls a route manifest, drives an optimized stop sequence, and captures proof-of-delivery (a scan, photo, or signature) at each stop, while the recipient tracks a live ETA and can redirect to a locker or reschedule. The three SLA tiers (same-day, next-day, scheduled) set the urgency that drives sort priority and route cutoffs.

Operational signals

Watch assign_latency_p95, stop_sequence_slip_minutes, manifest_queue_depth, pod_fail_rate, and geo_index_lag_seconds. Page when assign_latency_p95 > 12m for 5 minutes in any depot.

Why interviewers care

A strong answer treats this as a logistics-state problem: name Shipment Command as the single writer, separate the durable shipment state from the eventually-consistent route and ETA projections, and bring a real failure story — not a generic microservice diagram.

Interview checkpoint

Keep one concrete outage ready: a depot's geo-index lags, so assignments route associates to stale addresses — you catch it via geo_index_lag_seconds, freeze auto-assign for that depot, and fall back to last-known-good routes rather than mis-dispatching.