Design Uber ETA Prediction — System Design Practice | AmanAI Lab

Sheet›Design Uber ETA Prediction

AmazonGoogle

45:00

Hard

The Problem

Problem

Design Uber's ETA (Estimated Time of Arrival) prediction system that must predict pickup and trip times accurately for 20 million rides per day across 70 countries.

ETA predictions power:

•Driver-rider matching
•Surge pricing
•Route optimisation
•Customer notifications

What you'll be assessed on

Feature engineering for spatio-temporal data, model architecture for short vs long horizon ETAs, real-time traffic data integration, and evaluation metrics.

Scale & Constraints

▸20M rides/day, 200K concurrent trips at peak
▸ETA prediction within 100ms of ride request
▸P50 ETA error ≤ 2 minutes, P95 ≤ 5 minutes
▸Real-time traffic data updated every 30 seconds
▸Model accuracy must be maintained across 70 countries
▸Must handle special events (concerts, holidays) that break historical patterns

Must Cover0/9

Hints (if stuck)

💡 Segment the route into road segments and predict each segment's travel time independently, then sum.

💡 H3 hexagons (Uber's open-source library) partition the map for region-level traffic aggregation.

💡 Real-time traffic is the strongest feature — integrate it via a streaming join at inference time.

💡 Online learning: fine-tune on the last N=1000 trips in real time to catch emerging patterns.

0 words · auto-saved

Back to sheet

Mock interviewTake quizFlashcards

Problem

Design Uber's ETA (Estimated Time of Arrival) prediction system that must predict pickup and trip times accurately for 20 million rides per day across 70 countries.

ETA predictions power:

•Driver-rider matching

•Surge pricing

•Route optimisation

•Customer notifications

What you'll be assessed on