Hiring Manager / Technical Screening Real data engineering ownership Pipeline reliability understanding Product sense Communication clarity Depth vs buzzwords About OpenTable challenges: What do you think our main data challenges are? How would you measure data quality? How do you define tiered datasets? How do you store large-scale analytics data? Why lakehouse vs database? How would you design analytics for user behavior?

Coding Round (Python + SQL) Read logs line-by-line and count distinct error messages efficiently. Parsing logic Data structures (dict vs set) Handling large files Memory efficiency Clarity under pressure Follow-up: What if file is 10GB? How would you process at scale? Streaming vs batch? Avoid reprocessing same message? SQL Questions Q1: Report total party size booked per restaurant. Tested: JOIN GROUP BY SUM vs COUNT Alias usage Q2: For each restaurant, report quantity and revenue in its first booking year. Tested: MIN(year) per group Subqueries GROUP BY correctness Aggregation logic Joining back to bookings

System Design Round This was the deepest evaluation. Main Question: Design a pipeline that ingests, processes, and stores restaurant impression data at petabyte scale. They evaluated: A) Requirements clarification Logged-in vs logged-out? What metrics? What fields? Real-time vs batch? B) Architecture Messaging layer Storage format Bronze/silver/gold Partitioning strategy Processing tools C) Specific probing questions: What tools do you use to consume from Kafka? How do you avoid consuming the same message twice? How do you ensure idempotency? How do you enforce schema? What if producer changes data type? What format do you store in? How do you handle late-arriving data? What’s your Kafka partitioning strategy? How do you backfill historical data? How do you monitor pipeline health? This was testing: Streaming fundamentals Schema evolution Correctness guarantees Scale reasoning Tradeoffs Practical experience depth