Design a real-time system to maintain the remaining ETA for every driver within each cell of a city grid. You may only use the data provided in the prompt (no manual features or external data). Describe, step by step, how you would: ( 1) ingest and partition GPS pings and trip events; ( 2) denoise GPS signals and correct drift to robustly localize drivers within grid cells; ( 3) compute and continuously update each driver's remaining ETA; ( 4) select and train a model suitable for online serving; ( 5) perform online feature computation strictly from the approved data; ( 6) meet low-latency, high-throughput SLAs; ( 7) monitor, alert, and rollback models; and ( 8) evaluate offline and online. Specify concrete algorithms for GPS error reduction (e.g., filtering, map matching), data schemas, state stores and keys, windowing strategy, update frequency, failure handling, and your validation methodology.

This question evaluates a candidate's competency in real-time ML system design, covering streaming data ingestion, GPS noise handling and denoising, online feature computation, low-latency model serving, and operational monitoring.

How do I approach ML System Design interview questions?

ML System Design questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master ml system design interviews.

What difficulty level is this interview question?

This is a hard difficulty ML System Design question, commonly asked during Onsite rounds at Uber.

What role is this question designed for?

This question is commonly asked for Machine Learning Engineer candidates at Uber during technical interviews.

Design real-time grid ETA for drivers | Uber Interview Question

Real-Time Grid-ETA System Design

You are tasked with designing a real-time system that maintains the remaining ETA for every driver currently located within each cell of a city grid. You must use only the following data sources:

GPS pings from driver devices
Trip events (e.g., trip_start, trip_end, destination provided at trip start)

No external maps, POI data, or manual features are allowed. Assume:

The city is pre-partitioned into a fixed grid (e.g., S2/H3 or a rectangular grid) with a known cell_id function latlon → cell.
For an active trip, the destination (lat, lon) is known at trip_start.
GPS pings arrive at ~1 Hz with fields including timestamp and optional accuracy.

Design the system and describe, step-by-step, how you will:

Ingest and partition GPS pings and trip events.
Denoise GPS signals and correct drift to robustly localize drivers within grid cells.
Compute and continuously update each driver's remaining ETA.
Select and train a model suitable for online serving.
Perform online feature computation strictly from the approved data.
Meet low-latency, high-throughput SLAs.
Monitor, alert, and rollback models.
Evaluate offline and online.

Be specific about:

Algorithms for GPS error reduction (e.g., filtering, map matching)
Data schemas
State stores and keys
Windowing strategy and update frequency
Failure handling
Validation methodology

Real-Time Grid-ETA System Design

You are tasked with designing a real-time system that maintains the remaining ETA for every driver currently located within each cell of a city grid. You must use only the following data sources:

GPS pings from driver devices

Trip events (e.g., trip_start, trip_end, destination provided at trip start)

No external maps, POI data, or manual features are allowed. Assume:

The city is pre-partitioned into a fixed grid (e.g., S2/H3 or a rectangular grid) with a known cell_id function latlon → cell.

For an active trip, the destination (lat, lon) is known at trip_start.

GPS pings arrive at ~1 Hz with fields including timestamp and optional accuracy.

Design the system and describe, step-by-step, how you will:

Ingest and partition GPS pings and trip events.

Denoise GPS signals and correct drift to robustly localize drivers within grid cells.

Compute and continuously update each driver's remaining ETA.

Select and train a model suitable for online serving.

Perform online feature computation strictly from the approved data.

Meet low-latency, high-throughput SLAs.

Monitor, alert, and rollback models.

Evaluate offline and online.

Be specific about:

Algorithms for GPS error reduction (e.g., filtering, map matching)

Data schemas

State stores and keys

Windowing strategy and update frequency

Failure handling

Validation methodology

Design real-time grid ETA for drivers

Quick Overview

Real-Time Grid-ETA System Design

Solution

Comments (0)

Design real-time grid ETA for drivers

Quick Overview

Real-Time Grid-ETA System Design

Solution

Comments (0)