FDE Times: a 60-minute drill for the FDE technical screen — SQL, JSON and data checks in Python

Databricks-style Forward Deployed Engineer screens often last only 45–60 minutes over video, with many short questions and almost no time to look up syntax. FDE Times’s 4 November drill turns that into a 60-minute practice: four SQL patterns plus a short Python pass over dirty JSON.

The SQL half drills habits interviewers reward — and customers need in week one. COUNT(column) skips NULLs while COUNT(*) counts every row, so a NULL-rate query is the first data-quality check. ROW_NUMBER partitioned by business key (not blanket DISTINCT) dedupes when timestamps differ. For top-N, RANK and DENSE_RANK disagree on ties; the good answer is to ask how ties should be handled before filtering.

The Python half shows why a successful json.loads is not enough: by default the module accepts NaN and keeps the last value on duplicate keys, so bad lines never raise. A NaN self-inequality check plus an object_pairs_hook that rejects duplicate keys turns “3 good lines” into the real two-good / two-bad picture. Talk assumptions out loud on the video call; silence costs points even when the query is right.

Use the same three checks on a customer’s first export before you build a pipeline: NULL rates, duplicate counts by business key, and a trial JSON read that returns bad-line numbers. That is first-week FDE work, not only interview prep.