Big Tech platforms Data Engineer Interview Loop
Big Tech data engineer interviews cover SQL, Python and data modeling, in that order of how often candidates mention them. The screens differ. Meta splits an hour evenly between SQL and Python in a shared editor. Amazon runs either a 90-minute assessment or a live screen with a Leadership Principle question. The onsites look alike, with four to five rounds. Meta runs three "full-stack" interviews on one product scenario, plus an ownership round. Amazon runs a SQL deep dive, ETL, warehouse design, the hiring manager and the Bar Raiser. Algorithms count for less than in a software engineering loop, and data handling counts for more.
Practice by round
The basics in both languages, under time. In SQL that's joins, aggregations and distinct rows. In Python it's dictionaries, lists, counting, and the complexity of what you just wrote. Meta's screen sets a minimum number of correct answers in each half, so speed on the basics is the whole screen.
SQL at the level the loop asks it: window functions, CTEs, ranking, rolling averages, month-over-month change and time between events. The Python looks like pipeline work, such as flattening nested JSON, intersecting arrays, balancing frequencies, and finding a median efficiently. Interviewers also ask how the query behaves at scale and what you'd precompute.
| Question | Difficulty |
|---|---|
| Top 2 Users With Most Calls | Medium |
| Call Declines | Medium |
| Bottom 2 Companies By Mobile Usage | Medium |
| Company With Most Desktop Users | Medium |
| Signups By Billing Cycle | Medium |
| Daily Post Removal Ratio | Medium |
| Top Books by Checkout Duration | Medium |
| VR App Sessions | Medium |
| Car Part Price | Medium |
| Rank Variance Per Country | Hard |
| New And Existing Users | Hard |
| Most Popular Client For Calls | Hard |
| Top Actor Ratings by Genre | Hard |
| Search Click Success Rate by User Segment | Hard |
| Daily Revenue | Hard |
Start with the ideas a design conversation runs on: normalized versus denormalized tables, when a view saves storage and when it costs it, UNION versus UNION ALL, WHERE versus HAVING, and how a CTE runs. Then come design questions built on the companies' own products, like Friends You May Know, a restaurant recommendation and a database for customer addresses. The projects and mini-analyses are the end-to-end version: processing image files, resolving duplicate transactions and finding daily top users.
| Case study | Difficulty |
|---|---|
| Python Dictionary to Store Data | Easy |
| Are We Friends? | Medium |
| Computational Tricks | Medium |
| Grain of a Reactions Fact Table | Medium |
| Slowly Changing Seller Dimension | Medium |
| Order Header or Order Line: Which Grain? | Medium |
| Restaurant Recommendation | Hard |
| Late-Arriving Events in a Daily Engagement Table | Hard |