Data Engineering System Design Interview Questions
These are 100 design and data modeling questions that candidates reported from data engineering loops at 30+ companies between 2020 and 2026. Each row gives the company, the role, the year and a tag for how strong the evidence is. The questions are paraphrased and cut to the core ask.
The list is the appendix of Data Engineering System Design: Zero to Hero, where 26 of these prompts are worked into full designs.
Published 7 October 2026 · By Durgesh Yadav
Questions 1 to 40
MAANG + Microsoft (40)
Meta 11, Amazon 8, Microsoft 6, Netflix 6, Apple 5, Google 4.
Evidence in this group: 1 VERIFIED, 32 REPORTED, 7 PATTERN.
| # | Company | Role | Year | Question | Evidence |
|---|---|---|---|---|---|
| 1 | Meta | DE | undated | Round format, not a prompt: a product's data needs, then its logging, a data mart and reporting SQL. | VERIFIED3rd-party copy |
| 2 | Meta | DE | 2026 | Model a new housing-marketplace feature, define KPIs, then SQL for view-to-lead conversion by location. | REPORTED |
| 3 | Meta | DE | 2025, 2021 | Data model for a gaming company like Epic Games, with the schema trade-offs. | REPORTED |
| 4 | Meta | DE | 2026 | Model an Uber-style business, then KPI SQL, Python data processing and chart choices on the same model. | REPORTED |
| 5 | Meta | DE (Product Analytics), E4 | 2024-26 | Define success metrics for a product: ride matching, Twitter, a video platform, Reels, Marketplace. | PATTERN |
| 6 | Meta | Product DE, L5 | 2025 | Design facts and dimensions, state each relationship's cardinality, then SQL on a given schema. | REPORTED |
| 7 | Meta | DE, E4 | n/s | Defend a star schema and argue the pros and cons of the alternatives. | REPORTED |
| 8 | Meta | DE, Analytics | 2025 | How would you redesign your data model to make it more granular? | REPORTEDweak |
| 9 | Meta | DE | n/s | Model events that never happened, such as products never ordered, using factless fact tables. | PATTERN |
| 10 | Meta | DE | 2024-26 | Simulate a stream with lists and dicts: a running average, or ride requests per 15-minute window. | PATTERN |
| 11 | Meta | DE | 2025-26 | A core metric moves unexpectedly (a DAU spike, an Instagram drop): find the cause and the data needed. | PATTERNprep sites |
| 12 | Amazon | DE | 2026 | Design a data model for Amazon Prime from scratch; expect many follow-ups. | REPORTED |
| 13 | Amazon | DE | 2025 | Model retail from vendor to Amazon warehouse to customer; a separate round covered an AWS pipeline. | REPORTED |
| 14 | Amazon | DE | 2024 | Design a streaming data model and a batch one. Have you built SCD Type 2? | REPORTED |
| 15 | Amazon | DE | 2022 | ETL pipeline built for scale; model and query from scratch; ETL vs ELT; how you test integrity. | REPORTED |
| 16 | Amazon | DE II, L5 | 2021 | Design a data warehouse for a clothing store and write the DDL. | REPORTEDpre-2022 |
| 17 | Amazon | DE II, L5 | 2021 | Model your latest project, then slowly changing dimensions and surrogate keys. | REPORTEDpre-2022 |
| 18 | Amazon | DE, L5 | 2020-25 | Pipeline for a large dataset: justify the technology, explain how it scales and how you monitor it. | PATTERN |
| 19 | Amazon | DE, L5 | 2020 | Fact-dimension model: design the primary key and handle partial data. | REPORTEDpre-2022 |
| 20 | DE | 2026 | Real-time pipeline for clickstream from millions of users, such as ad-click analytics. | REPORTED | |
| 21 | DE | 2026 | Design and build an end-to-end batch data pipeline. | REPORTED | |
| 22 | AE (Data), L4 | 2024 | Central ERP integration that tracks warehouse inventory across a global network. | REPORTED | |
| 23 | AE (Data), L4 | 2024 | Car-rental reservations with vehicle tiers, discounts and multi-tenancy, down to the data model. | REPORTED | |
| 24 | Microsoft | DE | 2024 | Schema for employee check-ins and check-outs that yields hours worked, then query it. | REPORTED |
| 25 | Microsoft | DE | 2024 | E-commerce schema (products, orders, customers), SQL on it, then Spark performance tuning. | REPORTED |
| 26 | Microsoft | DE | 2024 | Data reconciliation and validation system; size Spark resources for different volumes. | REPORTED |
| 27 | Microsoft | Senior SWE (Data) | 2024 | Daily incremental S3-to-Oracle ingestion; estimate executor memory for 1 TB. | REPORTED |
| 28 | Microsoft | DE | 2024 | Design a real-time pipeline that processes streaming data. | REPORTED |
| 29 | Microsoft | DE | 2024 | Incremental loads in Azure Data Factory, normalization trade-offs, Z-ordering; redo a SQL answer in PySpark. | REPORTED |
| 30 | Netflix | Senior DE, L5 | 2025 | Rewrite a single-threaded Scala solution in Spark, then as a streaming job, weighing trade-offs. | REPORTED |
| 31 | Netflix | Senior DE, L5 | 2025 | Design a pipeline architecture, then work through debugging scenarios. | REPORTED |
| 32 | Netflix | Senior DE | 2026 | Design a data pipeline, then defend whether it scales to Netflix's volume. | REPORTED |
| 33 | Netflix | DE, L5 | 2025 | A real scenario from the team's own work; Spark depth and modeling of unstructured data. | REPORTED |
| 34 | Netflix | DE | n/s | Sketch the data or object model of a system you built; the round is fitted to the team. | PATTERN |
| 35 | Netflix | SWE, Senior+ | 2026 | Data model to track a direct-sold demand order on an advertising DSP. | REPORTEDweak |
| 36 | Apple | DE | 2024 | Data model for a given scenario, then medium-to-hard SQL on it. | REPORTED |
| 37 | Apple | DE | 2024 | Design a clickstream processing system for log data. | REPORTED |
| 38 | Apple | Analytics DE | 2025 | Design a pipeline, then a model that copes with schema evolution and slowly changing dimensions. | REPORTED |
| 39 | Apple | DE, Software DE | to 2025 | Spark depth: partitioning, skew, executor count and memory, dynamic allocation, MapReduce vs Spark. | PATTERN |
| 40 | Apple | Software DE, Media Products | 2025 | Scalable, reliable pipelines, and how the data feeds ML for personalized recommendations. | REPORTED |
Questions 41 to 70
US tech and finance (30)
Airbnb, DoorDash, Uber, Lyft, Stripe, LinkedIn, Pinterest, Snowflake, Databricks, Salesforce, Adobe, JPMorgan, Morgan Stanley, Robinhood.
Evidence in this group: 27 REPORTED, 3 PATTERN.
| # | Company | Role | Year | Question | Evidence |
|---|---|---|---|---|---|
| 41 | Uber | SWE | 2025 | Live driver heatmap: GPS pings into grid cells, density pushed to clients, top-K hottest cells per region. | REPORTEDaggregator |
| 42 | Uber | DE | 2025-26 | Rides or Eats data model serving city revenue, driver and merchant payouts, funnel and retention reports. | PATTERNprep sites |
| 43 | Uber | DE | 2025-26 | Real-time pipeline for driver location, matching or surge pricing with sub-second freshness. | PATTERNprep sites |
| 44 | Airbnb | Senior DE | 2024 | Design a bookings table and a properties table, then query them. | REPORTED |
| 45 | Airbnb | Senior DE | 2024 | A SQL dump lands in Hive daily partitions; from two given tables, write the ETL query for the target output. | REPORTED |
| 46 | Airbnb | DE | 2025 | Design several tables, choose their partitioning, explain scaling, then SQL on the model. | REPORTED |
| 47 | Airbnb | DE | 2021 | SQL that loads a new date into a ds-partitioned table from incremental data in earlier partitions. | REPORTEDpre-2022 |
| 48 | Airbnb | DE | 2026 | Model cumulative booking metrics plus each account's first and latest booking time. | REPORTED |
| 49 | Senior DE, IC3 | 2025 | End-to-end data system for several use cases: ETL, storage optimization, query performance. | REPORTED | |
| 50 | SWE | 2026 | Collect user activity events and answer counts for the last minute, hour and day. | REPORTEDaggregator | |
| 51 | Lyft | DE | 2026 | Design the driver and rider data models. | REPORTED |
| 52 | Lyft | DE | 2026 | Backend API that is the entry point for real-time driver request processing. | REPORTED |
| 53 | Lyft | DE | 2026 | Model a use case around a given query pattern; system design leaned toward a data platform. | REPORTED |
| 54 | DoorDash | DE | 2024, 2026 | Dimensional model, facts and dimensions, for a fitness app. | REPORTEDtwo mentions |
| 55 | DoorDash | DE | 2026 | End-to-end pipeline with both a batch and a real-time path. | REPORTEDaggregator |
| 56 | DoorDash | DE | 2026 | How do you handle late-arriving data? | REPORTED |
| 57 | DoorDash | DE | undated | Schema for a customer's address history over time, including past occupants and moves. | REPORTEDaggregator |
| 58 | Staff SWE (Data) | 2026 | Analytics platform that produces derived data; hot partitions and skew from popular boards. | REPORTED | |
| 59 | Staff SWE (Data) | 2026 | Top boards by engagement over a 5-minute sliding window; Kafka rebalancing and Spark shuffle internals. | REPORTED | |
| 60 | Stripe | SWE, Senior+ | 2026 | Count-metrics monitoring with rates, labels and alerts that copes with duplicate and late events. | REPORTEDaggregator |
| 61 | Stripe | SWE | 2025 | Near-real-time activity counter per user, device and region; offline clients sync later. | REPORTEDaggregator |
| 62 | Stripe | SWE, mixed | 2025-26 | Idempotent double-entry ledger, or an auditable ledger with reconciliation. | PATTERN |
| 63 | Snowflake | DE | 2025 | Keep a cumulative historical sales table up to date from current data, in SQL. | REPORTED |
| 64 | Snowflake | DE (per IQ) | 2026 | Metadata service for columnar storage: sharding, concurrent updates, min/max pruning, compaction. | REPORTEDaggregator |
| 65 | Databricks | DE (per IQ) | 2026 | Design a distributed file system without S3 or any commercial blob store. | REPORTEDaggregator |
| 66 | Salesforce | DE | 2026 | Your warehousing approach for bringing in data from many sources. | REPORTED |
| 67 | Adobe | DE | 2026 | End-to-end solution for a loosely defined problem; state assumptions, plan for scale. | REPORTED |
| 68 | JPMorgan | DE | undated | Architecture spec under hardware limits; requirements changed mid-presentation, so offer an alternative. | REPORTED |
| 69 | Morgan Stanley | DE | 2021 | Modeling and normalization, Databricks lakehouse, PySpark ETL design, batch vs streaming. | REPORTEDpre-2022 |
| 70 | Robinhood | SWE | 2026 | Distributed job scheduler with retries and SLA guarantees. | REPORTED |
Questions 71 to 100
India product and retail (30)
Flipkart, Razorpay, Walmart GT, Meesho, Paytm, Target, Nike, JioHotstar, Uber Hyderabad, VISA, Swiggy, unnamed startups.
Evidence in this group: 1 VERIFIED, 25 REPORTED, 4 PATTERN.
| # | Company | Role | Year | Question | Evidence |
|---|---|---|---|---|---|
| 71 | Razorpay | DE, 2.3 yrs | c. 2024 | A service that does Delta OPTIMIZE's job on S3 for a real-time workload, with monitoring and failover. | REPORTED |
| 72 | Razorpay | DE, 2.3 yrs | c. 2024 | Self-healing system that loads hundreds to thousands of SQL and NoSQL databases into a warehouse. | REPORTED |
| 73 | Razorpay | DE, 2.3 yrs | c. 2024 | Resume projects end to end: challenges, data quality metrics tracked, Spark internals. | REPORTED |
| 74 | Razorpay | DE, ~3 yrs | c. 2022 | SQL, data modeling and Spark internals in one round (topics, no prompt). | REPORTEDweak |
| 75 | Flipkart | DE II | 2024 | Model a cricket tournament where players turn out for several teams; SQL for cumulative player scores. | REPORTED |
| 76 | Flipkart | DE II | 2024 | Pipeline failure scenarios and consuming from Kafka in depth; Spark dynamic allocation, caching, joins. | REPORTED |
| 77 | Flipkart | DE-I to DE II | 2021-24 | Machine coding: join several (nested) JSON files in Spark and answer SQL-style questions. | PATTERN3 reports |
| 78 | Flipkart | DE-I | 2021 | Financial database: models, schema, fields, partition keys and query optimization. | REPORTEDpre-2022 |
| 79 | Flipkart | DE | 2021 | E-commerce data model with each table and relationship explained, then three SQL questions. | REPORTEDpre-2022 |
| 80 | Flipkart | DE | 2021 | Explain an ETL scenario end to end, from ingestion to the warehouse. | REPORTEDpre-2022 |
| 81 | Flipkart | DE | 2026 | Web crawling, API calls and anomaly detection, plus system design questions. | REPORTEDweak |
| 82 | Meesho | SDE-2 DE, ~3 yrs | c. 2022 | SQL, modeling and Spark internals, then Spark coding with Meesho's panel (topics, no prompt). | REPORTEDweak |
| 83 | Meesho | SDE III Data, 5-8 yrs | 2026 | Job description: fault-tolerant batch and streaming pipelines on Spark, Flink and Kafka; deep Spark internals. | VERIFIEDjob post only |
| 84 | Paytm | DE, 1 yr | 2023 | Design an online ticketing system: data flow and architecture. | REPORTED |
| 85 | Paytm | Tech Lead | 2025-26 | Design a data warehouse or a data lake. | REPORTEDtwo reviews |
| 86 | Walmart GT | DE-3 | 2023 | Design Mixpanel: capture events from Android, iOS and web apps. | REPORTED |
| 87 | Walmart | Senior DE | 2026 | End-to-end data platform: batch vs streaming, schema, storage layer; ingestion and recovery earlier. | REPORTEDtopics only |
| 88 | Walmart GT | DE to Senior DE | 2022-26 | Own design round; recurring topics: Scala Spark, Kafka, Airflow, Spark tuning, schema evolution. | PATTERN |
| 89 | Walmart GT | SDE | 2025 | Design an order schema, then an inventory management system. | REPORTEDSDE loop |
| 90 | Target India | DE | 2022 | What is a data pipeline and how would you build one? Warehouse vs lake. | REPORTED |
| 91 | Target India | DE | 2022 | Walk through an end-to-end project pipeline, then design and service comparisons. | REPORTEDweak |
| 92 | Nike | DE | 2021 | Bucketing external Hive tables, Parquet file sizing, bucket joins and map joins. | REPORTEDpre-2022 |
| 93 | Nike | DE | 2025 | Normalization, SQL optimization, partitioning and indexing. | REPORTEDweak |
| 94 | JioHotstar | SDE-2 DE | 2024 | Profile store merging daily attributes from many teams, each with a min, max, first or last rule. | REPORTED |
| 95 | JioHotstar | SDE-2 DE | 2024 | Design a threaded comment system like YouTube's. | REPORTED |
| 96 | Uber Hyderabad | SDE-2 DE, L4 | 2024 | Top-10 movies by category and window; a view counts at 80% watched, so merge overlapping intervals. | REPORTED |
| 97 | Uber Hyderabad | SDE-2 DE, L4 | 2024 | Clickstream pipeline for an hourly trending-by-location dashboard, scored on completed views. | REPORTED |
| 98 | VISA Bangalore | Senior DE | 2024 | Every step from a CSV file to a partitioned table; update 200 of 10,000 Hive rows with no timestamp. | REPORTED |
| 99 | Startups (unnamed) | DE, platform, SDE-Data | n/s | Distributed-systems prompts: a Kafka-like log, an event-driven system, a distributed cache. | PATTERNone thread |
| 100 | Swiggy | DE | 2026 | Deep dive on pipelines you ran at scale, framed on Swiggy order and delivery data. | PATTERNweak |
Chapter 01
What real loops asked, in 11 categories
The book sorts the 100 questions into these categories. The grouping is the author's own reading, and the last column lists the designs in the book that answer each one.
| Category | What was asked | Evidence | Designs in the book |
|---|---|---|---|
| A. Data ingestion | Razorpay c. 2024: self-healing loads from thousands of SQL and NoSQL databases. Microsoft 2024: an incremental S3-to-Oracle load sized for 1 TB. VISA 2024: a CSV to a partitioned table. | Good: several direct reports | 03, 20 |
| B. Lake / lakehouse | Razorpay c. 2024: a service that does the job of Delta OPTIMIZE on S3. Paytm 2025-26: design a lake or warehouse. Morgan Stanley 2021: a Databricks lakehouse. | Moderate: few full prompts | 11, 21, 24 |
| C. Warehouse / modeling | Meta 2026: model a housing-marketplace feature, then conversion SQL. Amazon 2026: Prime from scratch. Flipkart 2024: a cricket tournament, players in several teams. | Strongest: a verified guide plus most loops | 02, 12, 13, 14, 16 |
| D. Streaming | Google 2026: a real-time clickstream or ad-click pipeline. Uber Hyderabad 2024: hourly trending titles by location. Stripe 2026: counts that survive duplicate and late events. | Strong: many reports, some via aggregators | 01, 04, 08, 09, 10, 18, 19 |
| E. Batch | Google 2026: an end-to-end batch pipeline. Airbnb 2024: a SQL dump lands in Hive daily partitions, write the ETL. Amazon 2022: ETL vs ELT, and testing integrity. | Good | 12; batch paths in 04, 21 |
| F. Platform | Lyft 2026: a design round that leaned toward a data platform. Walmart 2026: an end-to-end platform and the storage choice. Robinhood 2026 (SWE): a job scheduler with SLAs. | Moderate: senior or SWE loops | 22, 23, 24 |
| G. Reliability | Netflix 2025: design a pipeline, then debug failure scenarios. Flipkart 2024: failures while reading Kafka. Stripe 2025-26: an idempotent double-entry ledger (a pattern). | Good, and the most common follow-up | 06, 08 |
| H. Data quality / governance | Microsoft 2024: a reconciliation and validation system. Razorpay c. 2024: the data quality metrics you tracked. Amazon 2022: testing integrity. No catalog or lineage prompt turned up. | Moderate | 15, 23 |
| I. Distributed systems | Databricks 2026: a distributed file system without S3. Snowflake 2026: a metadata service for columnar storage. Indian startups: build a Kafka-like log (a pattern). | Thin: aggregators and one forum thread | None of its own: Part 3, and rows 64, 65 and 99 |
| J. Analytics | Walmart GT 2023: design Mixpanel. LinkedIn 2026: activity counts for the last minute, hour and day. Meta 2024-26: success metrics for a product (a pattern). | Good | 05, 07, 09, 16, 17 |
| K. Business systems | Amazon 2025: vendor to warehouse to customer. Google AE 2024: inventory across a global warehouse network. Paytm 2023: an online ticketing system. | Good, but the domain changes each time | 02, 04, 05, 06, 13, 17, 25, 26 |
What the list says
Six things to take into the room
01
Modeling shows up in nearly every loop
Meta's guide gives modeling a round of its own, and candidates report modeling prompts at Meta, Amazon, Airbnb, Lyft, DoorDash, Flipkart, Microsoft and Apple in 2024-2026. Most then ask SQL on the model you drew, so a wrong grain costs you twice.
02
Late data and duplicates are the near-certain follow-up
DoorDash asked it outright in 2026. Stripe, LinkedIn and Airbnb list retries, double counts or late bookings, and Google's 2026 clickstream round covered both. Have the answer ready before you draw.
03
Requirements can change mid-round
A JPMorgan candidate had the requirements changed during the architecture presentation and had to offer an alternative on the spot. Writing your requirements on the board is what lets you see what changed.
04
Spark internals get tested next to design
Apple, Flipkart, Microsoft and Razorpay all went into executors, memory, skew, dynamic allocation or broadcast joins after the diagram. Expect "how many executors?" right after you draw.
05
Your past project is a design prompt
Netflix asks you to sketch the model of a system you built, and Razorpay, Target India and Amazon had candidates walk through a past pipeline. Bring one project already drawn and sized.
06
Some popular targets have almost no evidence
No data engineering design prompt turned up for Swiggy, Zomato, PhonePe or CRED. Google is thin for pure data engineering roles, and finance reports mostly list topics rather than prompts.
What this list cannot tell you
- Rows are not all independent: rows 75, 76 and 94-98 come from one candidate's 2024 write-up, rows 74 and 82 from another, and rows 30-32 may be one candidate. There are fewer independent reports than rows.
- Of the 84 REPORTED rows, 16 are weak snippets or older than 2022, and 8 US rows reach us through prep aggregators.
- No data engineering design prompt was found for Swiggy, Zomato, PhonePe, CRED, Zepto, Myntra or Ola, and no first-person Uber data engineering prompt. Finance loops gave topic lists or SWE-style low-level design.
- Some sources could not be read when the research was done in October 2026 (blocked, timed out or behind a login wall), so coverage has holes.
The worked answers are in the book
Data Engineering System Design: Zero to Hero
- 100 reported questions from 30+ companies, each tagged by evidence
- The PIPELINE method: eight steps for the 45 minutes
- 26 designs worked end to end, 19 cheat sheets and a master checklist
₹299 until 14 October, then ₹499. PDF, 111 pages.
More on the round: system design interview questions, system design prep by company and the system design mock interview.