The short answer: Applied scientists build models and algorithms for a company's products and help put them into production. Their interviews test ML depth and breadth, coding, and often a talk or deep dive on past work, plus behavioral questions tied to company values. Amazon publishes an Applied Scientist prep page with a phone screen and four 55-minute interviews. Uber publishes one Sciences interview guide shared by its data and applied scientists, including a 60-minute Jam Session it does not describe. Microsoft publishes only its general hiring process, so its applied scientist rounds come from candidate reports. Last reviewed October 2026.
Applied Scientist interview: quick answers
| Question | Short answer |
|---|---|
| What does an applied scientist do? | Builds models and algorithms for real product problems, tests them in experiments and works with engineers to ship them. Amazon and Microsoft postings stress putting models into production, and Uber says some of its scientists build models that run systems in real time. |
| What do the interviews test? | ML depth and breadth, coding, and a deep dive or talk on past work, plus behavioral questions. Uber adds experimentation and statistics, and candidate reports at Amazon and Microsoft describe ML system design. |
| Can I use AI tools? | Microsoft says no outside assistance unless explicitly permitted. Uber says no external assistance of any kind, AI tools included. Amazon publishes no rule on its careers pages, so ask your recruiter. |
| How much do applied scientists make? | Posted US base ranges run from $102.1K at Microsoft IC3 to $331.2K at Microsoft IC6 in the San Francisco Bay area and New York. Amazon lists $142.8K to $295.9K and Uber $161K to $211K. All three mention bonus or equity. |
| How is it different from data scientist, ML engineer and research scientist? | Amazon's Applied Scientist loop tests coding where its Research Scientist loop tests experimental design. ML engineers are engineering titles at all three. At Uber, applied and data scientists share one process and title ladder. |
| Is there an entry path? | Yes. Amazon runs Applied Science internships for PhD and Master's students, Microsoft posts Applied Scientist intern roles, and Uber lists Data Science & Applied Science Internships. |
What applied scientists do
Amazon says applied scientists solve practical problems that improve customer experience, cut costs and add automation. They develop models and algorithms, review the ML literature and work with software engineers on scalable implementations, and Amazon Science says they are heavily involved in putting their algorithms into production. US postings span Amazon Ads, Prime Video, AWS, Amazon Robotics and Stores.
Microsoft's Applied Scientists drive projects from design through experimentation to shipping, on teams such as Turing (Microsoft 365 Copilot), CoreAI, speech and translation, Bing web data and Microsoft Advertising. Some roles are not deep learning at all: the Ads auction team works on optimization, control theory and auction design. Every posting names an IC level, and pay bands are the same across disciplines at each level.
Uber's applied science work is mostly marketplace work: ML, optimization, causal inference and experiments for delivery matching, pricing, Uber One membership and Maps. Uber says some scientists use data to inform business decisions while others build models that run systems in real time. Its current postings use Scientist II and Senior Scientist titles with Applied Science as the subteam.
The neighbors differ by company. Amazon has a separate Research Scientist job family whose loop lists experimental design and data-driven modeling in place of coding, and it describes data scientists as generalists assessed on data skills, statistics and business acumen. Microsoft Research Researchers lead their own research agenda, and Microsoft Data Scientists focus on metrics, evaluation and A/B testing. At all three, Machine Learning Engineer is an engineering title, and Microsoft's and Uber's postings describe it building production ML systems with scientists. Uber publishes no explicit split between its applied and data scientists.
How 3 companies run the applied scientist interview
The short version of each process. Each company guide has the full detail, with links to the company's own pages, and says which parts come from candidate reports.
| Company | Screens | Main interviews | AI tools | What stands out |
|---|---|---|---|---|
| Amazon | One or two 60-minute technical phone screens | Four 55-minute interviews: science, coding, tech talk | No published rule; ask recruiter | Leadership Principles in every interview; Bar Raiser |
| Microsoft | 30-minute technical screen (candidate reports) | 2 to 4 conversations: coding, ML, design | Own skills only unless explicitly permitted | Same pay band across disciplines per IC level |
| Uber | 30-minute recruiter call, 60-minute live coding | 60-minute Jam Session, then 4 to 6 interviews | No external assistance, AI tools included | One process for all science roles |
The interview types in depth
ML depth and breadth
This is the core of most applied scientist loops. Expect questions across ML (losses, regularization, evaluation, model choice) and then deep follow-ups on the models you say you have used. Amazon lists knowledge depth and breadth as a loop competency, Uber lists Machine Learning, and candidate reports at Microsoft describe fundamentals such as precision and recall, ROC and AUC, and L1 versus L2.
A structure that works: give the definition, the assumption it rests on, how it fails and how you would detect that in real data. Tie each answer to a project where you made the choice. Interviewers go deeper until you reach the edge of what you know, so say plainly when you get there.
Coding
Applied scientists are expected to write production-quality code, and all three companies test it. Amazon lists problem solving and coding in its loop and links a recursive staircase sample. Microsoft says to code in your strongest language, with no pseudocode, in a tool that runs your code. Uber runs a 60-minute live coding session and lists SQL, Python and data structures and algorithms under Coding.
Practice medium problems in Python under time, plus small ML methods from scratch, such as k-means or a bag-of-words vectorizer. Talk through your plan before you write, then test corner cases out loud. For Uber, add SQL with joins and window functions.
Tech talk and project deep dive
Many loops spend a full round on your own work. Amazon lists a tech talk as a loop competency without describing it, and candidate guides describe a presentation of past research or projects to the panel. Candidate reports at Microsoft describe a deep dive on projects or research that includes how you would design ablations.
Pick one project with a real result and know it cold: the problem, the baseline, why you chose the method, the ablations, what failed and what you would do next. Keep your own contribution separate from the team's, and prepare a second project in case the first does not fit the team.
ML system design
You design an ML system end to end: the objective, data and labels, features, model, offline and online evaluation, serving and monitoring. A new-grad candidate report at Amazon describes ML system design rounds such as spam or fraud detection, and candidate reports at Microsoft describe RAG or end-to-end ML design. Microsoft also lists distributed systems topics for design prep.
Start by pinning down what the model's output drives and what a mistake costs. Then go through data, model and evaluation before infrastructure. Close with how you would detect drift and roll back a bad model.
Experimentation and statistics
Product-facing applied scientists are often tested on experiment design. Uber lists Experimentation/Statistics as a competency, and its postings ask for designing experiments and interpreting results, with causal inference and marketplace effects. Microsoft's technical page lists statistics such as hypothesis testing for data science areas.
Know sample size and power, how to read a result, and what breaks a simple A/B test. In marketplaces, treatment spills over to control, so be ready to propose switchback or cluster-randomized designs and to explain difference in differences.
Behavioral and values
Behavioral questions are scored against company values. Amazon asks two or three Leadership Principles questions in every interview and names Learn and Be Curious, Dive Deep, and Invent and Simplify for scientists. Microsoft suggests STAR(R) answers and focuses on growth mindset and customer obsession. Uber assesses every candidate against its values, such as Trip Obsessed and One Uber.
Write your main stories down with your own actions and a measured result. Science stories that land well include a model that failed in production, a result that did not hold up, and a time you chose a simpler method.
Pay in US job postings
US salary ranges from each company's own postings for this role, seen October 29, 2026. They are base salary unless the company guide says the posting lists total salary. Each company guide has the postings and links.
| Company | Titles in the postings | Posted US pay range |
|---|---|---|
| Amazon | Applied Scientist, Applied Scientist II, Applied Scientist III, Senior Applied Scientist and Principal Applied Scientist, plus Applied Science internships | $142.8K to $295.9K |
| Microsoft | Applied Scientist, Applied Scientist II, Senior Applied Scientist and Principal Applied Scientist, each posted with an Applied Sciences IC level (IC3 to IC6) | $102.1K to $331.2K |
| Uber | Applied Scientist and Sr Applied Scientist in its interview guide and older postings; current postings use Scientist II and Senior Scientist with the subteam set to Applied Science | $161K to $211K |
Entry-level paths
- Amazon: Applied Science internships for PhD students (3 to 6+ months, year-round starts), a 2027 track for Master's students in Seattle, and Postdoctoral Scientist roles.
- Microsoft: Applied Scientist intern roles for Bachelor's and Master's students, processed on a rolling basis from September to April, plus Microsoft Research internships and postdocs.
- Uber: Data Science & Applied Science Internships for university students and new graduates, listed on Uber's Emerging Talent page.
AI tools in applied scientist interviews
Rules on AI differ by company, so ask your recruiter before every round.
Amazon. Amazon does not publish a rule on AI use in interviews on its careers pages; ask your recruiter. Amazon says its new software development engineer assessment includes an AI assistant, but it ties that to SDEs, not applied scientists. A March 2025 news report described internal guidance asking candidates not to use generative AI tools during interviews unless explicitly permitted, which is a report and not a published Amazon rule.
Microsoft. Microsoft's Candidate Code of Conduct says: "During assessments and interviews, candidates should demonstrate their own skills without outside assistance unless explicitly permitted." It encourages candidates to use AI tools responsibly while preparing.
Uber. Uber's Sciences interview guide says: "Important: While we embrace AI in our work, do not use any external assistance during the interview process. This includes search engines, documentation sites, collaboration with others, or AI tools such as ChatGPT, Copilot, or similar."
A 6-week prep plan
| When | Focus |
|---|---|
| Week 1 | Read your target company's guide and postings. Note its title for the role, the rounds it publishes or candidates report, and its AI rule. Ask your recruiter which competencies and rounds you will get. Start the Machine Learning course. |
| Week 2 | ML breadth: losses, regularization, evaluation metrics, trees, linear models and neural networks. For each model on your resume, write down its assumptions and failure modes. Use the Machine Learning and Advanced Machine Learning courses. |
| Week 3 | Coding: medium problems in Python under time, plus k-means and a bag-of-words vectorizer from scratch, with tests. Add SQL for Uber. Use the Data Structures and Algorithms course. |
| Week 4 | ML system design: two problems a week, such as fraud detection and recommendations, from objective to monitoring. Use the ML System Design and Deep Learning courses. |
| Week 5 | Experimentation and your tech talk: sample size, A/B test readouts and marketplace designs with the Applied Statistics course, and a full rehearsal of a talk on your best project. |
| Week 6 | Behavioral and mock loops: map your stories to the company's values, then run two full practice loops with deep follow-up questions and no AI help. |
Courses for each round
- Machine Learning
- Advanced Machine Learning
- Deep Learning
- ML System Design
- Applied Statistics
- Data Structures and Algorithms
Practice questions
Questions from our bank, then practice prompts from the company guides.
Questions from our bank
From the DataInterview question bank, picked for applied scientist interviews. Open one to answer it.
Customer Lifetime Value
Amazon Sales Forecasting
Training with Few Labels
Calculating AUC
Transformer Decoder vs Encoder
Kmeans versus Gaussian Mixture
ETA Prediction Model
Ride Demand Forecast
Fraud Detection KPIs
Machine learning practice
One prompt per company, from the company guides. These are our prompts, not reported interview questions.
You are predicting whether a Prime Video ad impression leads to a purchase, and only 0.5% of impressions do. How do you train and evaluate the model?
Sample Answer
Use a probabilistic classifier trained with log loss, and judge it with PR AUC and calibration, since accuracy is useless at a 0.5% base rate. If you downsample negatives for speed, correct the predicted probabilities afterward. Split by time so the test set comes after training. Check calibration by segment, because ad decisions use the probability itself. Close with how you would confirm the offline gain in an online test.
Microsoft 365 Copilot answers questions over a user's documents. How would you evaluate answer quality before shipping a new model?
Uber's DeepETA predicts the residual between the routing engine's ETA and the actual arrival time. Why model the residual instead of the ETA itself?
ML system design practice
One prompt per company, from the company guides. These are our prompts, not reported interview questions.
Design a system that detects fraudulent seller accounts on Amazon's marketplace.
Sample Answer
Clarify the action (block, review, or limit) and the costs of a false positive versus a missed fraudster. Labels come from past investigations, which are delayed and biased toward cases that were reviewed. Use account, listing, payment and network features, with a fast model at signup and a richer batch model later. Route borderline scores to human review and feed outcomes back as labels. Monitor precision on reviewed cases and drift, since fraudsters adapt.
Design a retrieval-augmented generation system for Copilot that answers questions over a company's documents.
Coding practice
One prompt per company, from the company guides. These are our prompts, not reported interview questions.
Count the ways to climb n stairs when each step can be any size in a set X.
Sample Answer
Dynamic programming: ways[0] = 1 and ways[i] = sum of ways[i - x] for each x in X with x <= i. For n = 4 and X = {1, 2}, ways is 1, 1, 2, 3, 5, so the answer is 5. It runs in O(n times the size of X) time and O(n) space. Amazon's own sample question is a recursive staircase, so be ready to explain why plain recursion repeats the same subproblems many times and how memoization fixes it.
Implement k-means clustering from scratch in Python for a list of 2D points.
Experimentation and statistics practice
One prompt per company, from the company guides. These are our prompts, not reported interview questions.
Why can't Uber test a new driver pricing rule with a simple user-level A/B test? What would you do instead?
Sample Answer
Riders and drivers share one marketplace, so treating some drivers changes supply for everyone and the control group is affected too. Randomize by region and time instead: a switchback test flips the whole market between treatment and control in time blocks, or a cluster-randomized test assigns whole areas. Account for carryover between blocks and the smaller effective sample size.
Coding and SQL practice
One prompt per company, from the company guides. These are our prompts, not reported interview questions.
Write a SQL query that returns, for each city, the share of trips in the last 7 days that arrived more than 5 minutes after the ETA.
Sample Answer
Filter trips to the last 7 days, then GROUP BY city and take AVG(CASE WHEN actual_minutes - eta_minutes > 5 THEN 1.0 ELSE 0 END). Use 1.0 to avoid integer division. Exclude canceled trips or trips with a missing arrival time, and say so.
Behavioral practice
Prompts we wrote for applied scientist behavioral rounds, with what a strong answer covers.
Tell me about a model you shipped that did not work as expected in production.
Sample Answer
Describe the offline result, what happened online and how you found the cause. Explain what you changed in the model and in your evaluation process, and the result after the fix.
Tell me about a result that did not hold up when you checked it again.
Tell me about a time you chose a simpler method over a more advanced one.
Tell me about a disagreement with an engineer or product partner about a model.
Are you ready for applied scientist interviews?
1 / 5Can you explain the models you have used in depth: the loss, the assumptions and how they fail?
Company guides
Read the guide for each company you are interviewing with:
- Amazon: The Amazon Applied Scientist interview from Amazon's own prep page: the phone screen, the four-interview loop, the tech talk, the Bar Raiser and posted pay by title.
- Microsoft: The Microsoft Applied Scientist interview: what Microsoft publishes about its loop, what candidate reports add, the AI rule, and posted base pay from IC3 to IC6.
- Uber: The Uber Applied Scientist interview from Uber's own Sciences interview guide: five steps, the competencies, the no-AI rule, how the Scientist titles map, and posted pay.
Frequently Asked Questions
What is an applied scientist interview like?
Expect ML depth and breadth, coding, a talk or deep dive on your past work and behavioral questions. Some loops add ML system design or experimentation. Amazon and Uber publish their processes; Microsoft's applied scientist rounds come from candidate reports.
Do applied scientists have to code in interviews?
Yes, at all three. Amazon lists problem solving and coding in its loop, Microsoft has you code in your strongest language in a tool that runs your code, and Uber runs a 60-minute live coding session that includes SQL and Python.
Can I use AI tools in applied scientist interviews?
Microsoft says candidates should show their own skills without outside assistance unless explicitly permitted. Uber says no external assistance at all, including search engines, documentation sites and AI tools. Amazon publishes no rule on its careers pages, so ask your recruiter.
How much do applied scientists make at Amazon, Microsoft and Uber?
In US postings seen in October 2026, Amazon lists $142.8K to $295.9K base from Applied Scientist to Principal, Microsoft lists $102.1K to $331.2K from IC3 to IC6, and Uber lists $161K to $211K for Scientist II and Senior Scientist on its Applied Science subteam. All three mention bonus or equity.
What is the difference between an applied scientist and a research scientist?
At Amazon, Research Scientist is a separate job family whose loop lists experimental design and data-driven modeling where the Applied Scientist loop lists coding. At Microsoft, Applied Scientist postings stress shipping, while Microsoft Research Researchers lead their own research agenda. At Uber, Research Scientist titles appear on a 2022 Uber AI engineering blog post, separate from the Applied Science subteam.
What is the difference between an applied scientist and a data scientist?
At Amazon, Data Scientist is a separate job category of generalists assessed on data skills, statistics and business acumen. Microsoft Data Scientists focus on metrics, evaluation and A/B testing. Uber runs one interview process for both and separates them only by subteam, with no published split.
Do I need a PhD to be an applied scientist?
Not always. Amazon postings commonly ask for a PhD or a Master's plus several years of experience. Microsoft IC3 postings accept a Bachelor's plus 2 years, a Master's plus 1 or a Doctorate. Uber's Scientist II posting lists Ph.D., M.S. or Bachelor's degrees and asks for 4+ years of experience, or 2+ with a Ph.D.
