Amazon AI Research Scientist candidates report research and resume deep dives, ML and GenAI fundamentals, easy-to-medium coding, and Leadership Principles questions. Loop length and format vary by team.
$180K
Avg. Base Comp
$400K
Avg. Total Comp
2-6 rounds
Typical Rounds
Not reported
Process Length
Amazon AI Research Scientist candidates describe a loop that is broader than a pure research interview. The thread running through most accounts is that you need to explain your own work at more than one level of depth. Reports include a hiring-manager screen that was mostly a deep dive into a resume project, a first technical round that spent time on a candidate's PhD and a specific paper, and a tech talk with Q&A for a PhD student. One candidate was asked to describe current work in simple terms while the interviewer kept drilling into details, and another found a longer interview turned to mathematical expressions.
The ML questions mix fundamentals with modern GenAI. Candidates mention Transformer attention and time complexity, BERT versus GPT, DPO versus RLHF, and implementing a deep neural network, alongside classical topics such as K-means, linear regression, time series, and ROC/AUC. Coding is usually described as easy to medium: binary tree traversal, graph problems, a LeetCode-style array question, and a dynamic programming question. One candidate also had an online assessment with two medium-level coding questions, so don't assume the process will stay conceptual.
Leadership Principles appear in most accounts. One candidate had a Bar Raiser round made up entirely of them, and another was asked STAR-style questions in nearly every round, including influence without authority. Reported loop sizes varied, from a two-round virtual onsite up to a six-round virtual onsite spread over two days, and one candidate waited about two months just to reach the first screen. Treat team and level as variables and confirm the format with your recruiter.
Synthesized from 33 candidate reports by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the Amazon process.
Share your own interview experience to unlock all reports, or subscribe for full access.
Sourced from candidate reports and verified by our team.
Topics based on recent interview experiences.
Featured question at Amazon
Select the 2nd highest salary in the engineering department
| Question | |
|---|---|
| Merge Sorted Lists | |
| Experiment Validity | |
| Compute Deviation | |
| Weekly Aggregation | |
| Bagging vs Boosting | |
| Variable Error | |
| Button AB Test | |
| P-value to a Layman | |
| Prime to N | |
| Swipe Precision | |
| Nearest Common Ancestor | |
| Recurring Character | |
| Bank Fraud Model | |
| Jars and Coins | |
| Encoding Categorical Features | |
| Permutation Palindrome | |
| Radix Addition | |
| Find the First Non-Repeating Character in a String | |
| Valid Anagram | |
| Network Experiment Design | |
| Production Model Monitoring | |
| RMS Error | |
| Booking Regression | |
| Hurdles In Data Projects | |
| Dice Worth Rolling | |
| Random Bucketing | |
| Target Indices | |
| Success Measurement | |
| Swiping App Design |
Synthesized from candidate reports. Individual experiences may vary.
Candidates describe a recruiter conversation about background, interest in Amazon and the process, often followed by a screen with the hiring manager or a scientist. Screens varied: one was a deep dive into a resume project ending in a LeetCode medium DP question, one mixed ML basics with a binary tree problem, and one was mostly Leadership Principles with no coding.
Reported coding includes an online assessment with two LeetCode medium-level questions and a one-hour round with two graph problems. Some candidates found coding light, such as a round closer to LeetCode easy, an easier coding question in a manager interview, or K-means explained largely in pseudo-code. Not every candidate reports a dedicated coding round, so confirm the format with your recruiter.
Many candidates report a round built around their resume or research. Interviewers drilled into one project, discussed a PhD paper, or asked for a plain-language summary of current work before pressing on details. A clear high-level story followed by detailed tradeoffs is a sensible way to prepare.
Reported questions include attention and Transformer complexity, BERT versus GPT, DPO versus RLHF, implementing a deep neural network, ROC/AUC, time series, and Bayesian versus frequentist inference. Some loops added applied work, such as a case interview, ML system design for something like CTR, or an Amazon last-mile problem. Depth varies by interviewer, so prepare concise answers across the range.
Leadership Principles questions show up in most accounts, either in a dedicated Bar Raiser round or woven into several rounds. Reported prompts include influencing a decision without authority, handling conflict, and self-learning. Candidates recommend concise STAR stories you can adapt to different principles. Reported virtual onsites ranged from two to six rounds, with one spread over two days.
