D8LooP
Start building

Interviewing

Senior Data Engineer Interview Prep for 5–8 Years of Experience

Eight weeks shaped like a senior loop, where design and modeling decide the level.

8 weeks107 steps~33h 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 lead a design, defend one model, and tell ownership stories with real stakes.

Loop day Ready to be levelled senior

02

The plan, 8 weeks

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

  1. Week 01 · ~3h

    SQL at scale: correct, then cheap

    You can write interview-grade SQL and prove it is fast.

    Skip this if you tune a slow query by reading its plan, not guessing.

    1. SectionCommunicating trade-offs like a senior engineer7 minSeniors are graded on the trade-off, not the query.
    2. ExerciseNetflix: a 7-day unique count you cannot add up35 minin the studioRolling distinct counts: correct first, then affordable.
    3. ExerciseLate arriving event audit15 minin the studioLate events change yesterday. Find them before finance does.
    4. SectionReading a query plan7 minThe plan is your evidence. Opinions lose this round.
    5. ExerciseFind the predicates no index can help20 minin the studioPredicates that silently switch the index off.
    6. InterviewOptimizing over billions of rows6 minThe senior screen question. Start from bytes scanned.
    7. SectionPartition pruning, and what defeats it6 minPruning is the cheapest optimization you will ever ship.
    8. ExercisePartition pruning scan15 minin the studioProve the partition filter actually prunes.
    9. InterviewIndex vs partition6 minRow stores and warehouses answer this differently. Say which.
    10. SectionScan cost: bytes are dollars6 minPut a number on the query. Seniors are expected to.
    11. ExerciseInterview query-cost triage18 minin the studioRank the expensive queries from history by what they cost.
    12. ExerciseThe next page, without OFFSET18 minin the studioKeyset pagination, and why OFFSET falls over at depth.
    13. InterviewProving the optimization worked6 minBefore and after, measured. The follow-up they always ask.
    14. QuestionsSQL · Optimization & tuning15 minDiagnose from the plan, then name the cost.

  2. Week 02 · ~4h

    Modeling: commit to one design and defend it

    You can conform dimensions across processes and defend one trade-off.

    Skip this if you sketch a bus matrix and defend it under pushback.

    1. InterviewAnswering "design a model for X"6 minSeniors lead this conversation. Have the method ready.
    2. SectionConformed dimensions and the bus matrix8 minOne customer, one date, across every business process.
    3. SectionChoosing between the three fact types7 minTransaction, periodic or accumulating: argue it from the questions.
    4. ExerciseA hot dashboard over cold history50 minin the studioFast current state and full history in one design.
    5. ExerciseCurrent plan and monthly recurring revenue45 minin the studioMRR that survives upgrades, downgrades and mid-cycle changes.
    6. SectionThe medallion architecture6 minBronze, silver, gold, and what each layer promises.
    7. SectionThe semantic / metrics layer6 minOne definition of revenue, served to every tool.
    8. InterviewWide columns or a JSON payload6 minFlexibility against query cost. Pick one and defend it.
    9. SectionBitemporal modeling7 minWhat was true, and when we knew it.
    10. ExercisePolicies, endorsements, and mid-term changes55 minin the studioBitemporal history under mid-term changes. Senior-grade modeling.
    11. ExerciseCustomer 360 tradeoffs45 minin the studioOne customer from many systems, with honest trade-offs.
    12. SectionPartitioning and pruning huge tables6 minThe physical design decides the bill, not the diagram.
    13. InterviewHow do you review a model6 minSeniors review models too. Bring a checklist.
    14. QuestionsData Modeling · Approach trade-offs15 minState one trade-off per decision, then stop.

  3. Week 03 · ~5h

    Design I: batch, lakehouse and CDC

    You can size a batch lakehouse design before drawing it.

    Skip this if you open every design with numbers and a replay plan.

    1. SectionThe constraints nobody writes down6 minVolume, latency, freshness, cost: get them before boxes.
    2. InterviewCapacity is not just throughput6 minEstimate first. Skipping it is a named down-level reason.
    3. SectionLog-based CDC7 minReading the database log: the default senior answer for CDC.
    4. ExerciseMarketplace transactions at scale45 minin the studioMedallion at volume, with reprocessing designed in.
    5. SectionTable formats: Iceberg, Delta and Hudi7 minPick one for a stated workload, then defend it.
    6. InterviewCopy-on-Write vs Merge-on-Read6 minWrite cost against read cost. The workload decides.
    7. InterviewHow an update works on immutable files6 minWhat MERGE really does to your files.
    8. InterviewWhat compaction is, and when to run it6 minCompaction is a cost lever, not a chore.
    9. ExerciseThe source will not let you45 minin the studioRate limits, pagination, and resuming without starting over.
    10. ExerciseFourteen months of wrong numbers50 minin the studioFourteen months wrong. Rewrite it without taking the source down.
    11. ExerciseThe source that lies60 minin the studioIncremental extraction from a source that misreports its changes.
    12. ChapterSystem & Pipeline Design20 minHow design rounds are run, and what gets scored.
    13. QuestionsData Pipeline · Architecture design15 minClarify, estimate, design, then name what breaks.

  4. Week 04 · ~5h

    Design II: streaming, honestly

    You can design a stream with ordering, late data and effectively-once writes.

    Skip this if you can explain where exactly-once breaks and what fixes it.

    1. SectionPartitions: scale and ordering7 minOrdering is per partition. The key decides everything else.
    2. InterviewWhat ordering a stream gives you6 minThe promise is narrower than most candidates claim.
    3. SectionEvent time, processing time and watermarks8 minWatermarks trade completeness for latency. Say the trade.
    4. ExerciseAssign events to event-time windows40 minin the studioEvent-time windows by hand, late records included.
    5. ExerciseParcel tracking pipeline45 minin the studioOut-of-order events building derived state.
    6. SectionDelivery semantics in a stream7 minExactly-once means idempotent writes plus transactions. Qualify it.
    7. SectionPoison messages and dead-letter queues6 minOne bad record must not stop the stream.
    8. InterviewWhere the dead-letter queue goes6 minPlacement decides who can replay it.
    9. ExerciseRealtime events pipeline45 minin the studioThe canonical streaming design, end to end.
    10. ExerciseTen terabytes of clicks, and a question about sessions50 minin the studioSessions over terabytes: keyed state at scale.
    11. InterviewWhat backpressure actually is6 minWhat happens when consumers fall behind.
    12. ExerciseTen times the traffic, for two weeks50 minin the studioTen times the traffic: degrade on purpose, not by accident.
    13. InterviewOne stream or several?6 minA topology decision you justify by its consumers.
    14. QuestionsData Pipeline · Conceptual15 minSay the guarantee precisely, then its limits.

  5. Week 05 · ~5h

    Spark at senior depth

    You can fix skew, spill and sizing, and price the fix.

    Skip this if you size a cluster and design salting from first principles.

    1. SectionSalting, isolate-and-broadcast and AQE8 minNow you design the fix, not just spot the straggler.
    2. ExerciseOne user owns 30% of the events: which operations does a hot key hurt?25 minin the studioWhy AQE left this skewed join alone: read the thresholds.
    3. ExerciseSalting a skewed join key by hand: flatter tasks, same answer30 minin the studioDesign the salt, prove the result held, then question it.
    4. InterviewClassify the failure first6 minSymptom to cause in one pass. Seniors do not guess.
    5. SectionThe executor memory model7 minSpill is a design smell. Know where memory goes.
    6. ExerciseWhat a spill looks like, and how to make it go away25 minin the studioRemove a spill without adding memory. Partition size decides.
    7. InterviewExecution vs storage memory6 minThe unified pool, and what gets evicted first.
    8. SectionThe resource-sizing formula7 minCores, memory, executors: derive them, do not recite them.
    9. InterviewSize a cluster for 1 TB6 minWorked numbers, out loud. The senior sizing question.
    10. SectionReading the Spark UI7 minEvidence before tuning. The UI tells you which fix.
    11. SectionOptimising a slow job, step by step8 minOne slow job, fixed in order, each change measured.
    12. ExerciseFind the four suspects in a slow pipeline (one of them is innocent)30 minin the studioFour suspects, each convicted or cleared with a metric.
    13. SectionMERGE, schema evolution and maintenance7 minMERGE and OPTIMIZE: where lakehouse bills come from.
    14. SectionWatermarks, state and exactly-once7 minStreaming state grows until a watermark lets it go.
    15. InterviewFixing the small-file problem6 minCause, fix, and how to stop it recurring.
    16. InterviewChoosing a partition column6 minCardinality and query pattern decide. Argue both.
    17. ExercisePlan a small-file compaction40 minin the studioPlan a compaction: which files, merged into what.
    18. ExercisePlan which partitions a query must read45 minin the studioDecide which partitions a query can skip.
    19. QuestionsApache Spark · Performance tuning15 minEvery fix with its cost, and how you measured it.

  6. Week 06 · ~4h

    The senior signals: cost, failure and change

    You price a design, name its failures, and ship breaking changes safely.

    Skip this if you finished Senior / staff depth; these steps are shared.

    1. SectionSLIs, SLOs, SLAs and alerts people answer7 minSay the SLO before the architecture. Seniors own it.
    2. SectionThe failure modes you must expect7 minName what breaks before they ask.
    3. ExerciseYou fixed it. Who already used it?50 minin the studioYou fixed it. Now find everyone who used the wrong numbers.
    4. SectionHow to ship a breaking change anyway7 minExpand, migrate, contract: the answer to breaking changes.
    5. ExerciseIs this change safe?45 minin the studioProve a change is safe before it merges.
    6. ExerciseThe table everyone depends on45 minin the studioDeprecating a table other teams read, without an outage.
    7. InterviewThe source deleted a row6 minHard deletes upstream, and what history should say.
    8. SectionPaying for performance6 minWhich knob moves the bill, and by how much.
    9. ExerciseIt works, and it costs too much45 minin the studioIt works and costs too much. Cut it safely.
    10. InterviewOn-demand or provisioned capacity6 minA cost decision you should make with numbers.
    11. QuestionsData Pipeline · Cost & scale estimation15 minEvery design answer ends with what it costs.

  7. Week 07 · ~3h

    Ownership: the round that sets your level

    You can tell specific ownership stories with named people and business stakes.

    Skip this if you have five rehearsed stories naming stakeholders and outcomes.

    1. ChapterBehavioral Storytelling20 minSTAR, then make it specific: names, teams, numbers.
    2. ChapterAmazon20 minAt L6, Ownership and Dive Deep are probed hardest.
    3. ChapterMeta20 minThe Ownership round scores cross-functional alignment.
    4. InterviewTwo P1s at the same time6 minPrioritisation under pressure, told as a story you owned.
    5. InterviewWrong numbers, one hour to the meeting6 minStakeholder pressure. Show judgement, not heroics.
    6. SectionRunning the incident and writing it up7 minThe incident story they want: you ran it, then fixed the class.
    7. SectionThe best pipeline is often no pipeline6 minOver-engineering down-levels seniors too. Know when to stop.
    8. ExerciseWhen you cannot do it all60 minin the studioPartial data and hard trade-offs: decide, then explain.
    9. InterviewAirflow, Glue or Step Functions6 minTool choice defended with cost and team, not taste.
    10. InterviewMost common modelling mistakes6 minThe mentoring signal: name the mistake and how you teach it.
    11. ChapterFollow-Up & Negotiation20 minSenior offers are negotiated. Settle the level first.
    12. QuestionsData Pipeline · Trade-offs & tool choice15 minEvery choice with its reason and what it costs.

  8. Week 08 · ~4h

    Loop week: lead the design out loud

    You can run a full design round: clarify, estimate, design, fail, price.

    Skip this if you have run a timed mock design with a senior engineer.

    1. SectionDesign lab: a whole design, end to end8 minA whole design, clarifying question to trade-off.
    2. SectionNarrating a design6 minYou drive and the interviewer follows. Practise that order.
    3. SectionNarrating a modeling answer6 minGrain, conformance, history, cost: in that order.
    4. ExerciseListings, calendars and bookings55 minin the studioAn Airbnb-style marketplace model, against the clock.
    5. ExerciseCategory revenue that stays put45 minin the studioDimension history through a pipeline, as-of joins included.
    6. ChapterUber20 minStreaming-heavy design rounds. Batch-only prep fails here.
    7. ChapterAirbnb20 minA standalone modeling round, plus a metrics round.
    8. ChapterDatabricks20 minPlatform vendors expect opinions on Spark and Delta.
    9. ChapterSnowflake20 minSenior levels present a past project, then defend it.
    10. QuestionsData Modeling · Critique an existing model15 minCritique aloud: grain, keys, history, then cost.

03

Start it now

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

Start Senior Data Engineer · 5–8 yearsTracked, 8 weeks, ~33hSee the other plans16 more, by what you are aiming at