System Design Interview Practice

Reading a system design course teaches the vocabulary. Watching mocks on YouTube shows the rhythm. Neither produces the skill of designing under live pressure with follow-ups arriving mid-draw. The AI mock here runs that loop end to end: vague prompt, canvas, follow-up questions during drawing, mid-round requirement changes, and a structured verdict against the rubric.

Start a design mock Solo canvas first

Prepare for the interview

01 / Open invite

02min.

Know the patterns before the interviewer asks them.

a system design query, the same shape a screen would give you.

The diff against expected. Where ties broke. What you missed.

sandbox

1source → bronze → silver → gold

2 ingest : CDC + Kafka

3 transform : dbt + Airflow

4 serve : Snowflake

Execute your solution0.4s avg.

PayPalInterview question

Solve a problem

30 min

Per mock

Prompt families

Seniority levels: mid, senior, staff

Unlimited

Free mock sessions

How a 45-minute mock unfolds

Timeline from prompt to verdict. The mid-round requirement change at 18:00 is the differentiator between mid-level and senior performance.

Anatomy of a 45-minute mock design session

  0:00 ─── PROMPT
        "We get 40M clickstream events/day. Dashboard refresh
         every 15 min. Design the pipeline."

  0:30 ─── CLARIFY (you should ask)
        - schema of the events?
        - cost constraint?
        - dashboard backend?
        - retention?  ◄── AI may answer vaguely on purpose

  2:00 ─── PROPOSE (1-2 sentences before drawing)
        "Land events to S3 hourly, dbt incremental into
         Snowflake every 15 min, dashboard reads gold tables."

  3:00 ─── DRAW on canvas ◄── AI interrupts: "why dbt over Spark?"
                              "what about late events?"
                              "where does cost live?"

 18:00 ─── REQUIREMENT CHANGE
        "Latency requirement just tightened to 1 min."
        ◄── do you redesign cleanly or freeze?

 28:00 ─── DEFEND the redesign

 38:00 ─── STRETCH from AI
        "What if events grow 10x?  Would you change anything?"

 44:00 ─── VERDICT
        per-dimension score + decisive moments + drill suggestions

Transcript excerpt from a senior CDC round

The follow-up arriving while you're still drawing is the part most candidates underestimate. The mid-round requirement change is what staff rounds test specifically.

Mock CDC design round, senior rubric, transcript excerpt

AI: You need to replicate a production Postgres to a Snowflake warehouse. Replication lag under 5 min. Design it.

You: What's the schema change cadence? And is there a sustained write rate I should design around?

AI: Schema changes quarterly. Writes peak at 8K TPS, normal 2K.

You: [drawing] Debezium reads the WAL, publishes to Kafka, schema registry, sink connector to Snowflake.

AI: Why Debezium over logical decoding direct?

You: Debezium handles snapshots and resume from offsets; doing logical decoding direct means I'd build the same machinery.

AI: Quarterly schema changes. Which step breaks if a column is renamed?

You: The sink connector. Schema registry catches it but the downstream Snowflake DDL has to follow. I'd add a migration step that detects schema diff and runs the rename.

AI: [mid-round change] The product team wants a 60-second freshness floor now. Does anything in your design change?

You: Kafka and Debezium are fine at sub-minute. The bottleneck moves to the sink connector batch size and Snowflake's micro-partition write rate. I'd switch to Snowpipe streaming.

Verdict highlight:Senior-pass. Schema-evolution handling was the decisive moment. Pivot to Snowpipe streaming on requirement change scored at senior level.

What the rubric weights at each seniority level

Same 6-dimension rubric. The weights shift with level.

Level

Rubric weight shifts to

What the AI listens for

Mid-level

Correctness 30% · Communication 25% · Edge cases 20%

Did you commit to a specific tool? Did you name a freshness number? Did you respond to the first follow-up coherently?

Senior

Tradeoff articulation 25% · Failure modes 25% · Cost 15%

Did you surface the tradeoff before being asked? Did you name late-data, replay, or schema drift relevant to the prompt? Did you state a cost band?

Staff

Adapt on fly 25% · System thinking 25% · Implicit requirements 20%

Mid-round requirement change handled cleanly? Recognized what the prompt didn't say but a designer must address? Anticipated the next 2 questions?

4-week prep arc with mock cadence

Volume of mocks, solo canvas attempts, and reading per week. The arc compounds: vocabulary first, solo practice, then mocks weighted to weakness.

4-week design interview prep arc, with mock cadence

Week 1Build vocabulary (RPO, RTO, exactly-once, AQE, watermarking)

Mocks: 0Canvas: 1Reading: 5

Week 2Solo canvas problems, iterate on rubric verdicts

Mocks: 1Canvas: 6Reading: 2

Week 3AI mocks weighted on weakest dimension from prior verdicts

Mocks: 3Canvas: 4Reading: 1

Week 4Mocks + peer mock + 1 dry-run with a friend

Mocks: 3Canvas: 3Reading: 1

System design mock FAQ

Is this a course or a tool?+

A tool. The page is the practice surface: prompt, canvas, follow-ups, verdict. No lessons to complete. Pair with a course (ByteByteGo, Educative's Grokking, HelloInterview) if you need the patterns first. Read those to absorb, use this to produce under pressure.

How long is a typical session?+

30-60 minutes per mock including verdict review. Mid-level prompts run shorter (30-40 min); staff-level with mid-round changes run longer (45-60 min). The 4-week arc above shows the recommended cadence.

Does AI mock replace a peer human mock?+

Not entirely. Better for volume, rubric consistency, and immediate verdicts. Worse at social pressure simulation. Run mostly AI mocks during prep, layer in 1-2 peer mocks (Pramp, interviewing.io, a friend) in the last 2 weeks for the social rep.

What if I'm new to system design?+

Start with reading. 2 weeks on ByteByteGo or Educative's Grokking course to absorb the vocabulary. Then start canvas in solo mode (no mock interviewer). The first 3-5 verdicts will be rough; the trajectory steepens after that.

How does this differ from SWE system design prep sites?+

Different prompts and different rubric. SWE design (DesignGurus, Tech Interview Handbook) leans on request-path systems: design Twitter, design Instagram. DE design leans on pipelines: design CDC, design clickstream-to-dashboard. The vocabulary overlaps but the scoring differs.

Do I need to memorize specific cloud services (Kinesis, Dataflow, Data Factory)?+

Recognize them, not memorize. The interview interest is the pattern (managed streaming vs self-managed Kafka), not the service name. The canvas includes major managed services per pattern. Memorizing service-specific configuration is rarely tested.

02 / Why practice

Start a senior CDC mock

01
Active recall beats re-reading by 50%
Cognitive-science meta-reviews (Dunlosky et al., 2013) rank practice testing as a top-tier study technique, while re-reading and highlighting rank near the bottom
02
76% of hiring managers reject on the coding task, not the resume
From HackerRank's 2024 Developer Skills Report. Candidates who look strong on paper still fail the live screen if they haven't done timed, executable practice
03
Five problem shapes cover 80% of data engineer loops
Dedup, sessionization, top-N-per-group, slowly-changing dimensions, partition tricks. Writing the shapes by hand turns the unfamiliar into pattern recognition

Start the mock

Related practice

Solo Canvas Problems→

Same canvas, no interviewer, for pattern-building before mocks.

Pipeline Architecture Mock→

DE-specific framing of the same mock interview.

DE System Design Guide→

Round breakdown, the rubric in detail, decision framework.