Uber Data Engineer Interview Questions
The short answer
Uber's data engineering loop tests your ability to handle massive, real-time marketplace operations. You will face a technical screen followed by five onsite rounds covering SQL deep dives, Python data manipulation, dedicated Kimball data modeling, streaming-heavy system design, and behavioural scenarios. Expect questions on event-driven architectures, geospatial partitioning, and complex state-machine trip lifecycles.
A guide to navigating Uber's real-time data engineering pipeline, from SQL deep dives to streaming system design.
Or skip ahead and have your answer scored — free, no account needed.
The Uber Data Engineer interview process
Uber's data engineering loop is highly structured around real-time operations and high-throughput pipelines. Unlike other tech companies, Uber isolates data modeling into its own rigorous session and heavily prioritises streaming paradigms over batch-only processing. You must demonstrate a deep understanding of event-driven architectures and marketplace dynamics.
Typically 3-5 weeks from recruiter screen to offer
Recruiter Screen
30-minute phone call
- Basic role fit and compensation expectations
- Location and visa requirements
- High-level alignment with Uber's operational scale
Technical Phone Screen
60-minute video call via CoderPad
- Code execution speed
- Syntax correctness in SQL and Python
- Foundational reasoning regarding data manipulation under load
For junior L4/SDE II levels, this round may occasionally shift to a standard LeetCode style Data Structures and Algorithms (DSA) or Spark-basics screen.
SQL Deep Dive
45-to-60-minute live coding on CoderPad
- Execution of complex window functions
- Handling date boundaries and event stream sessionisation
- Writing optimised multi-table joins and computing marketplace KPIs
Coding Round (Python)
45-to-60-minute live coding
- Ability to write clean, maintainable, production-quality data manipulation logic
- Flattening complex JSON objects and API parsing with retries
- Handling streaming data state
This round does not grade abstract, dynamic programming puzzles.
Data Modeling Round
45-to-60-minute virtual whiteboard session
- Ability to build a complete star schema dimensional model from scratch
- Identifying correct grain and designing fact and dimension tables
- Handling Slowly Changing Dimensions (SCD) and low-latency schemas
System Design (High-Level Design)
45-to-60-minute virtual whiteboard session
- Knowledge of streaming versus batch paradigms with Kafka and Flink
- Scalability trade-offs and fault tolerance
- Geohashing or geospatial partitioning using H3
What Uber grades across the whole loop
- Streaming-first architectural reasoning
- Dimensional modeling fluency using Kimball methodologies
- Operational ownership under stress
- Real-time marketplace KPI computation
9 Uber Data Engineer interview questions
These are drawn from what data engineer candidates report being asked at Uber. Under each one is what Uber is testing and what a strong answer actually contains — not what technique to use.
SQL & Python Coding
These rounds evaluate your ability to write clean, optimised code under load, focusing on real-time marketplace metrics and data transformation.
“Write a query to find drivers who didn't take a trip in their first 7 days.”
Why they ask it
Tests your ability to handle date boundaries and compute activation metrics across user cohorts.
What a good answer contains
Correctly handling date arithmetic without hardcoding, using left joins or subqueries to isolate non-events, and optimising for large-scale user tables.
“Given a stream of Uber ride requests, compute the number of ride requests in every 15-minute window.”
Why they ask it
Evaluates how you handle event stream sessionisation and time-series data manipulation.
What a good answer contains
Implementing sliding or tumbling window logic in Python, handling out-of-order events, and managing state efficiently without memory leaks.
“Write a function to transform nested JSON data into a flat structure suitable for warehouse ingestion.”
Why they ask it
Tests production-quality data manipulation and parsing of complex, semi-structured API payloads typical of ride events.
What a good answer contains
A recursive Python function that handles nested dicts and lists, maintains clean schema mapping, and includes error handling for malformed JSON.
Data Modeling & System Design
Uber isolates data modeling into a dedicated round and expects system designs to default to streaming architectures.
“Design the data model for Uber's ride business. We need to support city-level revenue dashboards, driver earnings reports, and weekly cohort retention for riders.”
Why they ask it
Tests your ability to build a Kimball star schema that supports both low-latency operational reporting and deep historical analytics.
What a good answer contains
Defining the exact grain of a ride event, designing fact and dimension tables, handling Slowly Changing Dimensions (SCD) for driver statuses, and modeling the complex lifecycle of a trip.
“Design an architecture to compute real-time surge pricing across cities.”
Why they ask it
Evaluates your knowledge of high-throughput streaming systems, geospatial partitioning, and low-latency state tracking.
What a good answer contains
Utilising Kafka for event ingestion, Flink for real-time aggregation, H3 geohashing for spatial partitioning, and Pinot for low-latency querying.
“Design a clickstream pipeline to power a trending dashboard.”
Why they ask it
Assesses your ability to build scalable, fault-tolerant ingestion pipelines that handle massive event streams.
What a good answer contains
Explaining the trade-offs between lambda and kappa architectures, detailing watermark latencies, and ensuring exactly-once delivery semantics.
Behavioural & Collaboration
These questions assess your alignment with Uber's core values, particularly extreme ownership and customer obsession.
“Tell me about a time a data pipeline broke in production. What did you do?”
Why they ask it
Tests your reaction to high-stress scenarios and your ability to perform structural mitigation.
What a good answer contains
Detailing the exact blast-radius metrics, immediate mitigation steps, root-cause analysis, and the long-term engineering fixes implemented to prevent recurrence.
“Tell me about a time you owned a data quality issue that was not strictly your responsibility.”
Why they ask it
Evaluates your alignment with Uber's 'Go get it' value of extreme operational ownership.
What a good answer contains
Describing how you identified a silent data degradation, collaborated across teams to trace it to an upstream source, and established permanent monitoring.
“Tell me about a time you made a technical decision with incomplete information.”
Why they ask it
Assesses how you balance velocity and risk in a fast-paced operational environment.
What a good answer contains
Explaining the trade-offs considered, how you gathered proxy data, and how you built a rollback plan to mitigate potential negative impacts.
A worked answer, with the structure showing
This is written to be spoken, not read. Do not memorise it — take the shape and put your own experience through it.
The question
“Tell me about a time a data pipeline broke in production. What did you do?”
Senior Data Engineer, three years at a logistics and delivery platform.
Situation — 15 seconds
At my last company, our core driver payout pipeline failed on a Friday afternoon due to an unannounced upstream schema change that dropped a critical geohash field.
Action (Triage) — 20 seconds
I immediately calculated the blast radius—affecting payouts for 12,000 drivers—and rolled back the consumer schema to the previous version to stop the failing runs, while setting up a temporary fallback to default coordinates.
Action (Resolution) — 25 seconds
I coordinated with the upstream product team to restore the field, reprocessed the dead-letter queue using a backfill script I wrote, and verified that all driver balances were correct before the nightly payout run.
Outcome & Prevention — 20 seconds
We met our payout SLA with zero financial discrepancy. To prevent this, I implemented automated schema validation at the ingestion layer and set up Slack alerts for schema drift.
Why it works at Uber
- Quantifies the blast radius immediately, showing operational awareness of business impact.
- Demonstrates hands-on triage and backfill execution under tight SLAs, aligning with Uber's 'Go get it' value.
- Focuses on structural mitigation like schema validation at ingestion rather than just a one-time fix.
Free · no account needed
Say your answer out loud. Get it scored.
Reading about the structure is not the same as saying it. Answer “Tell me about a time you owned a data quality issue that was not strictly your responsibility.” the way you would in the room, and get a score plus three specific fixes. Aim for 60-90 seconds.
This browser cannot record audio. .
Practise these out loud with AI feedback
Reading the questions is not the same as answering them. Paste the actual Uber job posting and your CV, and prepare.fyi builds the twenty questions that loop is most likely to ask — then you answer them out loud, with a live AI interviewer and a scored breakdown of every answer.
What candidates get wrong in this loop
Modeling Trips as Static Events
Candidates often treat an Uber trip as a static row in a database. In reality, a trip is a complex state-machine lifecycle (request to match to cancellation or completion) with split payments, geographic boundaries, and multiple rating systems.
Fix: Always model the transitions and state changes of a trip, incorporating event-driven state tracking rather than just the final state.
Assuming Batch-Only Architectures
Relying solely on Spark, Hive, or dbt batch processing paradigms. Uber's platform demands real-time dispatch and surge pricing.
Fix: Address watermark latencies, exactly-once delivery, and Kafka state tracking in your system design rounds to show streaming competency.
Lack of Dimensional Modeling Fluency
Freezing up during star schema generation on whiteboard tools because of a lack of practice.
Fix: Practise drawing out star schemas in real time using digital canvas tools like Miro or Excalidraw, focusing on Kimball methodologies.
Uber Data Engineer interview: frequently asked questions
- How long does the Uber Data Engineer interview process take?
- The end-to-end process realistically takes 3 to 5 weeks. This includes 1 to 2 weeks for the initial screens, a 1-to-2 week gap to coordinate panel availability for the virtual onsite, and 3 to 5 business days for the final decision post-onsite.
- Is there a dedicated data modeling round at Uber?
- Yes. Unlike most tech companies that fold modeling into system design, Uber has a standalone 45-to-60-minute Data Modeling round. It focuses strictly on Kimball dimensional modeling, star schemas, and handling Slowly Changing Dimensions (SCD).
- What streaming technologies does Uber expect candidates to know?
- Uber heavily prioritises streaming architectures. You should be familiar with event-driven patterns using Kafka and Flink, and understand why Uber built and open-sourced tools like Hudi, Pinot, and the H3 geospatial indexing system.
- What is the format of the Uber Data Engineer technical screen?
- The technical screen is a 60-minute video call via CoderPad. It is typically split into 30 minutes of Python coding and 30 minutes of SQL. For junior L4 levels, this may occasionally shift to a standard LeetCode DSA or Spark-basics screen.
Other roles at Uber
Uber
Software Engineer interview questions
Backend and general engineering loops: coding, system design, and behavioural rounds.
Uber
Frontend Engineer interview questions
UI building under time pressure, JavaScript depth, and frontend system design.
Uber
Product Manager interview questions
Product sense, metrics and analytics, prioritisation, and execution rounds.
The questions every Uber round opens with
Each has the structure, example answers, and the same free grader.
Go deeper
How this page was put together
Compiled from public candidate reports, Uber’s own published material, and interview write-ups, last checked 6 August 2026. Interview loops change and vary by team, level and office — treat this as a strong prior, not a script. If something here no longer matches what you were sent, tell us and we will correct it.