Finished the KPMG data engineer interview loop last week for a position on their managed services data platform team. There's almost nothing online about this specific track so I'm writing it up.
A bit of context: this role is a step inside consulting, meaning I'd be building and maintaining data pipelines for client engagements, not internal tools. That framing matters for understanding what they tested.
The full loop for my DE role: Recruiter screen: 30 min logistics Technical screen: 60 min, mix of SQL and Python Onsite virtual: two rounds, one pipeline architecture, one behavioral
Technical screen breakdown:
SQL (about 35 minutes): Three problems of escalating complexity. Window functions, multiple CTEs, and one aggregation with a tricky GROUP BY. The hardest question involved calculating rolling averages over non-uniform time intervals. They used a shared SQL editor, I think it was their own internal tool.
Python (about 20 minutes): One question. Given a list of JSON records representing API responses, parse and transform them into a flat tabular format, handle nulls and nested keys. Not algorithmic, very much a real-world data wrangling task. I used pandas.
Onsite pipeline architecture round:
Here's where KPMG gets interesting for DE. The prompt was: design a data pipeline that ingests client transaction records from three different source systems (two batch, one near-real-time), normalizes them, applies validation rules, and loads to a data warehouse. It had to handle schema drift.
They pushed me on: error handling strategy, schema-on-read vs. schema-on-write tradeoffs, how I'd handle PII in transit, audit logging for compliance. Very Big 4 flavored.
Tools I mentioned: dbt for transformation, Apache Airflow for orchestration, Snowflake as the warehouse. They seemed comfortable with this stack. They also mentioned their internal clients use Azure heavily so I mentioned Azure Data Factory compatibility.
Verdict on difficulty: SQL was genuinely challenging, Python was practical, architecture was more about communication and tradeoff reasoning than knowing the 'right' answer.