Amazon Data Scientist interview questions

182 Data Scientists have shared notes from their Amazon interviews, with the questions they were asked in each round. That is 12% of the 1,581 Amazon interview notes on Screna AI.

182 Data Scientist interview notes

Amazon·Data Scientist·Technical Phone Screen

Jul 2026
SQL technical screen for a Data Scientist role at Amazon. One question, pretty focused, centered on window functions and how to pull first-per-group records out of an orders table.
  • Given an orders table with merchant IDs, order IDs, amounts, and dates, write a SQL query that returns each merchant's first order for every calendar date they had activity.

“I knew ROW_NUMBER() was the right move but fumbled for a second on whether to partition by just merchant_id or by both merchant_id and order_date.”

View Post

Amazon·Data Scientist·Onsite - Behavioral / Leadership

Jul 2026
Behavioral loop at Amazon for a Data Scientist role, focused almost entirely on how you handle things going sideways. Four questions, all circling the same theme: pressure, prioritization, and keeping your team from falling apart.
  • Tell me about a major risk or crisis that came up in the middle of a project and how you handled it quickly.
  • When you're working with limited resources, how do you decide what to cut or deprioritize without dropping the ball on delivery?
  • Walk me through how you run a post-mortem after something goes wrong. How do you make sure the same mistake doesn't happen again?
  • How do you keep your team motivated and productive when the project is under sustained pressure for a long stretch?

“I had a decent story ready but fumbled the 'rapidly' part.”

View Post

Amazon·Data Scientist·Technical Phone Screen

Jul 2026
Amazon Data Scientist interview that went deep into ML fundamentals, covering ensemble methods, regularization, transformer internals, and gradient issues all in what felt like one very long breath. Technically dense and not super conversational.
  • Compare Random Forest and XGBoost in terms of bias-variance trade-off, training speed, and interpretability.
  • What is overfitting, and what are at least three techniques you would use to reduce it in a large-scale recommendation system?
  • What evaluation metrics would you consider for a recommendation model, and when is each one preferable?
  • Explain how LoRA adapts large transformer models, and contrast CNN, RNN, and Transformer architectures including why attention helps with long-range dependencies.
  • What causes gradient vanishing and exploding, and how do batch normalization, residual connections, and careful weight initialization each help mitigate those problems?

“I knew this one pretty well but still fumbled the interpretability angle a bit.”

View Post

Amazon·Data Scientist·Technical Phone Screen

Jul 2026
Amazon data scientist interview with a case-style question about diagnosing a business decline. Pretty open-ended, which I wasn't fully prepared for.
  • If a key business metric suddenly drops, how would you go about diagnosing the root cause using data? Walk through the metrics you'd look at and where you'd get that data.

“I started with conversion funnel breakdowns and felt okay about that, but I fumbled when they pushed on external factors.”

View Post

Amazon·Data Scientist·Technical Phone Screen

Jul 2026
Amazon data scientist interview that was basically a deep dive into a single gnarly fraud modeling scenario. One question, four parts, and an hour to convince them you actually know what you're doing with imbalanced data in production. Felt more like a take-home crammed into a live session.
  • You've inherited a fraud detection model with severe class imbalance (positives around 2%) and weak validation performance. How would you diagnose whether it's underfitting or overfitting, propose three interventions you could ship in a single day, design a threshold selection procedure based on a specific cost structure, and define the minimum monitoring you'd put in place at deployment?

“This was a lot to hold in your head at once.”

View Post

Amazon·Data Scientist·Technical Phone Screen

Jul 2026
Amazon data scientist interview that was heavier on SQL internals than I expected. Most of the session was about cross-database compatibility and query semantics, the kind of stuff you think you know until someone asks you to explain it out loud.
  • A MySQL query using a column alias inside HAVING works fine, but the same query breaks on MS SQL Server. Why does this happen, and how would you rewrite it to work on both databases?
  • What does SELECT * FROM A B; do, and why is it valid SQL?
  • Among ORDER BY id, ORDER BY id ASC, and ORDER BY id DESC, which produces a different result from the other two and why?
  • Write the correct skeleton for a query that defines two CTEs and then selects from both of them.
  • For a column containing only uppercase letters, will these three predicates return identical rows: col IN ('A','B','C'), col BETWEEN 'A' AND 'C', and col >= 'A' AND col <= 'C'? Justify your answer.

“This one stung a little because I actually use MySQL day to day and never thought about why the alias worked.”

View Post

Amazon·Data Scientist·Technical Phone Screen

Jul 2026
Amazon Data Scientist technical screen, one meaty SQL problem that covered window functions, retention logic, and revenue ranking all in one shot. The question was well-constructed and felt like something you'd actually write on the job, which made it slightly more stressful than a pure puzzle.
  • Given a game_sessions table and an ad_impressions table, write a SQL query that returns one row per user with their first play date, a 7-day retention flag, total ad revenue in the first 7 days from their first session, their country, and a dense rank of users within each country by that revenue figure.

“This one had a lot of moving parts and I almost forgot to handle the case where a user has zero ad impressions.”

View Post

Amazon·Data Scientist·Technical Phone Screen

Jul 2026
Amazon data scientist interview that went deep into imbalanced classification metrics. The whole session was basically one long applied ML problem with four parts, and the math gets uncomfortable fast if you haven't touched precision-recall curves recently.
  • Given 12 scored examples with a severe class imbalance (about 1% positive rate), compute precision, recall, and F1 at thresholds of 0.90, 0.60, and 0.50.
  • Which threshold maximizes F1 on this dataset, and why might business cost considerations push you toward a different threshold anyway?
  • Explain when PR curves are more informative than ROC curves, and what a comparison of AUPRC versus AUROC would tell you in an extreme class imbalance setting.
  • If you can only surface exactly k=2 alerts, compute precision@k and recall@k for the top-2 scored examples, and explain how model calibration affects your ability to use a fixed score threshold in practice.

“The arithmetic itself isn't bad but I kept second-guessing which examples fall above each threshold boundary.”

View Post

Amazon·Data Scientist·Onsite - Behavioral / Leadership

Jul 2026
Amazon behavioral loop for a Data Scientist role. One question, but it was a beast, basically five questions stitched into one with very specific expectations around dates, artifacts, and process changes.
  • Walk me through a specific time you missed a deadline and a separate time you went beyond your formal responsibilities. For each: what were the exact dates and measurable impact, what trade-offs did you consciously make, which stakeholders did you loop in and when, what did your recovery plan look like with leading indicators you tracked, and what process changes would have prevented the problem or accelerated the positive outcome?

“This question wrecked me a little.”

View Post

Amazon·Data Scientist·Onsite - System Design / Architecture

Jul 2026
Amazon Data Scientist loop with a combined system design and coding round. Two pretty different problems back to back, which made the whole thing feel a bit disjointed, but I got through it.
  • Design a distributed order-processing system that is both highly available and low-latency. Walk through the architecture, how you'd scale it, and what fault-tolerance mechanisms you'd put in place.
  • Implement LeetCode 987, vertical order traversal of a binary tree, and explain the time and space complexity of your solution.

“I started with a pretty standard event-driven setup, message queue in the middle, separate services for inventory, payment, and fulfillment.”

View Post