D8LooP
Start building

Interviewing

Data Engineer Interview Prep in 3 Weeks

Three weeks shaped like the loop, weighted by what 56 companies actually reject people for.

3 weeks49 steps~10h total
Start this plan

It opens on your dashboard and ticks itself off as you go. Every step below is a chapter or an exercise that already ships.

01

When you finish, you can

You can pass a SQL screen, survive the Spark round, and defend a design out loud.

Loop day You can answer the round you were losing

02

The plan, 3 weeks

Ordered by what each one buys you, not by topic. Every step says why it is here.

  1. Week 01 · ~3h

    The screen: SQL, fast and correct

    You can take a prompt to a defended answer without freezing.

    Skip this if you already pin the output contract before writing any SQL.

    1. SectionClarify, contract, build, verify, explain7 minFive moves. The candidates who freeze are the ones improvising.
    2. SectionGrain, filters, time, dedupe, output7 minMost wrong answers are wrong here, not in the syntax.
    3. InterviewIN vs EXISTS6 minThe opener almost every screen uses. The speed answer is folklore.
    4. ExerciseInterview fanout-safe order revenue20 minin the studioFan-out is the most-asked SQL trap. Prove your total.
    5. InterviewDISTINCT vs GROUP BY6 minThe same fan-out, asked as a definition. DISTINCT usually hides it.
    6. ExerciseInterview top buyers with ties18 minin the studioTies are where a nearly-right answer gets marked wrong.
    7. SectionTop N per group6 minTop-N per group is on the screen in almost every loop.
    8. ExerciseInterview latest snapshot status18 minin the studioLatest row per key. The most reused window pattern there is.
    9. ExerciseInterview cohort retention slice25 minin the studioCohorts and retention: what product-facing teams ask.
    10. InterviewDay 1 / 7 / 30 retention6 minThe definition is the hard part. They ask which one you mean.
    11. SectionEdge-case checklist6 minNULLs and empty groups. Rejection notes name these by name.
    12. InterviewReconciling two systems6 minAsked at BitGo: ledger against settlement, and what each side is missing.
    13. InterviewN events within a rolling window6 minAsked at BitGo: three failed withdrawals inside any ten minutes.
    14. InterviewWhy is this query slow?6 minFive buckets. Have the structure before they ask for the plan.
    15. QuestionsSQL · Query formulation15 minTwenty prompts. Say the contract before you write anything.
    16. QuestionsSQL · Output prediction15 minThe query runs. Say what it returns without running it.

  2. Week 02 · ~3h

    Spark: the round that fails most people

    You can read a Spark plan, name the bottleneck, and fix skew.

    Skip this if you can already diagnose a shuffle from the Spark UI.

    1. InterviewThe three components, named6 minThe opening question of most Spark screens. Most candidates name two.
    2. InterviewHow many DAGs does a script build?6 minThree actions, three jobs, three recomputes. Asked to check you looked.
    3. SectionAnatomy of a shuffle7 minEvery Spark performance question is a shuffle question underneath.
    4. SectionReducing & avoiding shuffles6 minWhat you can actually do about it, in order of leverage.
    5. ExerciseAggregate before the join or after it? Let the shuffle sizes decide25 minin the studio"Aggregate before joining", tested with and without a broadcast.
    6. SectionDetecting skew in the Spark UI6 minOne task still running after the other 199 finished.
    7. SectionSalting, isolate-and-broadcast & AQE7 minSalting, broadcast, AQE. Know which one the question wants.
    8. ExerciseOne user owns 30% of the events: which operations does a hot key hurt?25 minin the studioEven count, skewed join, silent AQE. Explain all three.
    9. InterviewWalk me through fixing skew6 minAsked as a story. Diagnosis first, cheapest fix first, then the number.
    10. SectionReading the Spark UI7 minReading the UI aloud answers "how would you debug this".
    11. SectionBottleneck signatures, configs & small files7 minBottleneck signatures, and the configs that actually move them.
    12. ExerciseIf AQE coalesces anyway, does spark.sql.shuffle.partitions still matter?20 minin the studioThe config everyone tunes, and what AQE already does with it.
    13. InterviewClassify the failure first6 minSeven classes. Reading the error for its class bounds the search.
    14. InterviewSize a cluster for 1 TB6 minThe numerical round. Subtract the overhead out loud before you divide.
    15. InterviewWhy small files hurt, step by step6 minEveryone says small files are bad. Say what Spark does per file.
    16. QuestionsApache Spark · Performance tuning15 minTwenty tuning questions. Say the signature, then the fix.
    17. QuestionsApache Spark · Conceptual & architecture15 minThe internals round. "Shallow" is the logged rejection here.

  3. Week 03 · ~4h

    Design it, then say it out loud

    You can design from requirements and tell the story behind it.

    Nothing to skip — this is the round most candidates are eliminated in.

    1. SectionThe requirements-first method7 minRequirements first. Naming tools first is the documented failure.
    2. SectionThe questions to ask first6 minWhat to ask before you draw a single box on the whiteboard.
    3. InterviewChoosing a partition key6 minOne choice sets ordering, distribution, hotspots and scale. Say all four.
    4. ExerciseRealtime events pipeline45 minin the studioA prompt lifted from real loops: batch or streaming, and why.
    5. InterviewBatch to streaming, safely6 minThe architecture is the easy half. They ask about the cutover.
    6. InterviewAnswering "design a model for X"6 minWorkload, grain, tables, split. Naming tables first loses the marks.
    7. InterviewDesign a financial ledger6 minAsked at BitGo. Double-entry, or an unbalanced transaction stays possible.
    8. SectionChoosing the right SCD type6 min"Wrong SCD, unjustified schema" is a logged rejection, verbatim.
    9. ExerciseCategory performance over time40 minin the studioReport over a dimension that changed. Defend the type you picked.
    10. ExerciseIt has to be there by six45 minin the studioThe deadline round. What you drop, and what you never drop.
    11. InterviewCopy-on-Write vs Merge-on-Read6 minDelta and Iceberg are on the posting. This is what gets asked.
    12. SectionNarrating a design6 minNarrating a design is a separate skill from having one.
    13. InterviewTwo P1s at the same time6 minThe on-call round. Ranking two incidents out loud is the whole test.
    14. ChapterBehavioral Storytelling20 minA vague story with no measurable impact is a logged rejection.
    15. QuestionsData Pipeline · Architecture & design15 minDesign prompts, out loud. Requirements before components.
    16. QuestionsData Modeling · Critique an existing model15 minHere is a model. Say what breaks, and when.

03

Start it now

Reading is free and needs no account. The plan itself tracks on your dashboard.

Start Interview in 3 weeksTracked, 3 weeks, ~10hSee the other plans16 more, by what you are aiming at