Data Engineer Mock Interview

Window functions, CTEs, joins, aggregations, and subqueries executed against a real database. The AI checks correctness, edge case coverage, and query performance. Every question runs your actual SQL, not multiple choice. Topic breakdown: Window functions 34%, Joins 28%, Aggregation 22%, CTEs 10%, Subqueries 6%

Data manipulation, ETL logic, file parsing, API handling, and pandas. Code executes in a sandbox with access to pandas, numpy, and the standard library. The grader evaluates correctness, edge case handling, and adherence to Pythonic patterns. Topic breakdown: Data manipulation 40%, ETL logic 25%, File parsing 15%, API handling 10%, Pandas 10%

Star schema design, fact vs dimension tables, slowly changing dimensions, data vault, and medallion architecture. The AI interviewer asks follow-up questions based on your answers, exactly like a real discussion round. Topic breakdown: Star schema 30%, SCDs 20%, Normalization 20%, Data vault 15%, Medallion 15%

Batch vs streaming, idempotency, orchestration, failure handling, and schema evolution. System design rounds where the AI probes your reasoning with follow-up questions that get harder as you answer correctly. Topic breakdown: Batch vs stream 25%, Reliability 25%, Orchestration 20%, Schema evolution 15%, Scaling 15%

Partitioning strategy, shuffle optimization, broadcast joins, UDFs, and Spark SQL. Questions target the performance tuning and debugging that interviewers care about, not just API syntax. Topic breakdown: Partitioning 30%, Joins 25%, Performance 20%, SQL 15%, Streaming 10%

Most DE interview pools cycle through roughly 40 core question patterns dressed up as hundreds of variations. The candidate who recognizes that a 'find consecutive events' prompt is a gaps-and-islands problem can solve it in around 12 minutes. The candidate who brute-forces it runs out of time. The bank groups questions by pattern so candidates can build recognition rather than chasing variants.

A query that produces the right output on sample data leaves several questions open: does it handle NULLs, does it survive duplicate timestamps, does it return the right answer on an empty table. Interviewers at Google and Amazon allocate meaningful partial credit based on the count of edge cases a solution handles. The AI grader reports which cases passed and which broke, with the specific failing input.

Internal scorecards at most large companies separate problem solving from communication and weight them comparably. Explaining the approach before coding, naming a tradeoff out loud, walking through the rationale for one design over another all contribute. The discussion-mode rounds in the simulator exercise this directly with follow-up questions that score the depth of the response, not just the initial answer.

A mid-level candidate may solve a medium SQL problem in 25 minutes. A senior candidate solves the same problem in 10 minutes and uses the remaining 15 to discuss optimization, edge cases, and scaling behavior. The simulator tracks solve time per question and per topic; topics where solve time runs long are flagged for additional drilling.

The question pool is built from interview reports submitted by candidates who interviewed at Meta, Google, Amazon, Netflix, Uber, Airbnb, Stripe, Snowflake, Databricks, and roughly 265 other companies. Questions target the most frequently observed patterns rather than offering a random selection; each question maps to a specific pattern documented in the underlying dataset.

The questions and scoring rubrics on the platform are written with input from data engineers who have conducted interviews at large tech companies, using the structure of internal scorecards as a reference. Hints, evaluation criteria, and follow-up prompts reflect how interviewers score candidates in practice rather than the more generic feedback most coding platforms provide.

Questions are tagged Easy, Medium, and Hard based on the time and depth observed in real interview rounds rather than academic complexity. A Medium question targets 15 to 25 minutes of solve time, matching the per-question budget in a 45-minute coding round after accounting for intros and discussion. A Hard question targets the senior and staff bands.