The short answer
Complexity should mean moving parts + clear ownership, not buzzword stacking.
How to structure the answer (5–7 minutes)
- 1Business goal: who consumes the data
- 2Sources: DBs, APIs, files, events
- 3Architecture: extract → land → transform → serve
- 4Hard parts: late data, SCD, joins, quality, retries
- 5Orchestration / monitoring
- 6Your role: what you personally built
Example outline
> Goal: daily customer 360 for marketing. Sources: orders (Postgres), events (S3 ), CRM CSV. Landed raw to , staged with schema checks, transformed with Spark/SQL into dim_customer (SCD2) and fct_orders. Orchestrated in Airflow with sensors on file arrival. Hard parts: late events and duplicate order IDs; handled with watermark cutoff and merge on business key. I owned the staging + SCD2 model and the row-count / uniqueness tests.
Fresher-honest advice
A solid 3-stage project pipeline beats inventing Kafka + Flink + multi-region. Depth over theater.
Interview tip: Draw boxes mentally: Source → Lake/Warehouse → Mart → Dashboard. Mention failure modes.