Prepare for your Machine Learning and Data Science interviews
Created by a staff ML scientist at a leading unicorn startup (formerly at Meta), MAD Interview is a new comprehensive platform for Machine Learning and Data Science interview preparation.
-
As a guest
Explore 50 machine learning and data science interview questions with detailed answers—no login required.
-
With your Gmail
Access 68 machine learning and data science interview questions and answers by logging in with a gmail account.
-
Go Premium
Support my effort by paying a small fee to access 175 machine learning and data science interview questions and answers.
-
As a guest or with your Gmail
Access a role-specific (i.e., product data science, data science, applied science, or machine learning engineering) subset of questions and answers that define the lowest bar to succeed in these interviews.
-
Go Premium
Support my effort and access the complete set of role-specific (i.e., product data science, data science, applied science, or machine learning engineering) questions and answers that define the lowest bar to succeed in these interviews.
-
With your Gmail
Run your own SQL queries on a real database schema with real data. Follow a smal subset of a structured class that teaches you everything you need to know about SQL for product data science and data science interviews. (to be completed in 2025)
-
Go Premium
Support my effort to access the complete structured class that teaches you everything you need to know about SQL for product data science and data science interviews. (to be completed in 2025)
-
As a guest, with your Gmail, or with Premium
Read unfiltered tips on building your resume and understanding the application and interview process for different roles within machine learning and data science. I share my personal experience on application success rates, referrals, negotiations, and more.
Track your progress
MAD provides a detailed dashboard that allows you to track your progress across different topic areas (e.g., machine learning, machine learning coding, probabilities, statistics, SQL, A/B testing, etc.). Choose your target role and let MAD identify your gaps and guide you towards preparing to succeed in your interviews.
Login| Topic | Questions & Answers |
|---|---|
| Machine Learning | 78 |
| Statistics | 20 |
| Probability theory | 16 |
| SQL | 15 |
| A/B Testing | 13 |
| Data Structures and Algorithms | 9 |
| Machine Learning Coding | 9 |
| Application process | 8 |
| Metrics | 4 |
| Technical deep dive | 2 |
| Behavioral | 1 |
Role-specific and topic-specific questions with answers
Easy, medium, and hard questions that cover most topics in machine learning and data science interviews. Solutions that deep dive into explaining complicated concepts, with necessary references and simulations when needed.
Designed with rigor by an active interviewer
As a staff ML scientist, I regularly interview candidates at all levels—from new graduates to staff—while actively contributing to my company's question bank. Previously, I conducted interviews for product data science roles at Meta.
My backgroundFocus on learning
The bar book collection focuses on all practical topics that a data/applied scientist or machine learning engineer need to master, not only to pass the interview but also to succeed on the job.
Detailed solutions
I dedicate several hours to each question to craft the most thorough solutions possible. When necessary, I include references for further reading. I believe this rigorous approach sets MAD apart from other platforms.
Run your code
I put a lot of effort into designing test cases that you can test your machine learning code (e.g., implementing SGD, k-means, and more). I have only seen this before in actual interviews or take-home tests. I believe that such instant validation that your code is correct is an integral compoenent of learning and building confidence.
The interview process
Learn what to expect in a data/applied science or machine learning engineering interview loop across different levels. Learn from my mistakes, and get insights from personal statistics across referrals, timelines, success rates, negotiations, and more.
Pricing
I am keeping subscription prices extremely low, especially as I roll out new features throughout 2025. Your support is greatly appreciated.
Gmail
FreeThis option is best if you are just exploring the material of this platform.
Login to start- 68 interview questions with answers
- Run your code
- Run SQL queries on real data
- Access a sample of the bar books
- Track your progress
- Learn about the interview process
Almost ready
$5per month, paid every two months. This option is best if you have already started interviewing and have passed a few initial rounds.
Login to subscribe-
175 interview questions with answers
- Run your code
- Run SQL queries on real data
- Access the complete series of the bar books
- Track your progress
- Learn about the interview process
- Support my effort and give me energy to continue improving the platform
Starting now
$4per month, paid every four months. This option is best if you are fmiliar with the role you are applying to but haven't started interviewing yet.
Login to subscribe- 175 interview questions with answers
- Run your code
- Run SQL queries on real data
- Access the complete series of the bar books
- Track your progress
- Learn about the interview process
- Support my effort and give me energy to continue improving the platform
New grad or new role
$3per month, paid every six months. This option is best if you are a new grad or you are transitioning from a different role to a ML/DS role.
Login to subscribe- 175 interview questions with answers
- Run your code
- Run SQL queries on real data
- Access the complete series of the bar books
- Track your progress
- Learn about the interview process
- Support my effort and give me energy to continue improving the platform
Frequently asked questions
I created MAD as a side project out of my own frustration during transitioning from academia to industry and while preparing for interviews. A typical ML (DS) interview loop includes:
- Coding ML (Coding SQL)
- ML System Design (or metrics, root cause analysis, problem-solving)
- Coding Data Structures and Algorithms (A/B testing)
- ML Breadth and Depth (probability, statistics, ML concepts)
- Behavioral, leadership, and previous experience
Despite this structure, most online resources focus primarily on coding (Data Structures and Algorithms and SQL), which only accounts for about 20% of the interview process. Some interview preparation platforms do offer incomplete preparation in probability, statistics, A/B testing, and ML concepts. However even these platforms often do not provide solutions or use solutions from user-generated content, which can be inaccurate and misleading.
Additionally, there is no concentrated, role-specific learning material for candidates. Typically, candidates have to refer back to their college notes, take multiple courses on Coursera, download slides from CS229, and randomly skim through probability/stats and A/B testing books. MAD aims to consolidate all this into role-specific online books, covering the fundamental knowledge candidates need to brush up on before their interviews. Below is an outline of the areas MAD aspires to cover, making it the go-to platform for Machine Learning and Data Science interview preparation. If you have any comments or suggestions, please email me at comments@madinterview.com.
Below I summarize the differences between MAD and other existing platforms.
| Topic | Data Structures and Algorithms (DSA) Interview-prep Platforms | Data Science Interview-prep Platforms | MAD (expected in 2025) |
|---|---|---|---|
| Coding, Data Structures and Algorithms | Best coverage | Some coverage | Curated list of the type of DSA questions that can be asked in ML/DS interviews (20+ questions and answers) |
| Coding, SQL | Some coverage | Good coverage | Curated list of SQL questions that cover all concepts asked in Product Data Science and Data Science interviews (20+ questions and answers) plus an SQL playground with a real database schema to explore |
| Coding, Machine Learning Concepts | Do not cover | Some coverage | Extensive coverage (20+ frequently seen in interview ML concept coding questions and answers) |
| Coding, Machine Learning on Real Data | Do not cover | Coverage of take-home case studies | Coverage of freqently-seen hour-long interview sessions for applied science/MLE roles (5+ solved ML modeling problems from real interviews) |
| Probability and Statistics | Do not cover | Good coverage, but often without answers | Extensive coverage, broken down by topics, with proper solutions and references (50+ questions and answers) |
| A/B Testing | Do not cover | Some coverage, but often without answers | Extensive coverage (20+ questions and answers) |
| Machine Learning | Do not cover | Some coverage, but often without answers | Extensive coverage (100+ questions and answers) |
| Machine Learning System Design | Do not cover | Some coverage, but often without answers | Framework for preparing on how to answer these questions plus system design knowledge questions (10+ questions and answers, 2+ system design problems) |
| Metrics and Root Cause Analysis | Do not cover | Some coverage, but often without answers | Framework for preparing on how to answer these questions plus metrics-specific knowledge questions (5+ questions and answers, 2+ root cause analysis problems) |
| Behavioral and Prior Experience | Do not cover | Do not cover | Framework for preparing on how to answer these questions (20+ questions to prepare) |
| Focus on Learning | Limited | Limited | Extensive coverage of the fundamental knowledge that a candidate needs to have before interviewing (50+ relevant topics) |
| Role-specific Interview Prep | Do not cover | Do not cover | Interview-prep material curated for Product Data Science, Data Science, Applied Science, and Machine Learning Engineer roles |
Please let me know if you think MAD is falling short in any of these categories. I am always looking for feedback to improve the platform.
No I am not an influencer -- in fact I am quite an introvert and would never be able to become an influencer.
I am currently a Staff Applied Scientist at a unicorn startup and have previously held roles as a Staff (IC6) Research Data Scientist and later Staff (IC6) Product Data Scientist at Meta. I have years of experience conducting data science interviews and creating relevant interview questions. I hold a PhD in applied ML and an undergraduate degree in Engineering. Before Meta, I spent seven years as a researcher and academic, teaching data science at a top-tier liberal arts college in the northeast US. My diverse background and academic experience allowed me to interview for and receive offers for MLE, Applied Scientist, Data Scientist, and Product Data Scientist roles from top-tier companies such as Google, LinkedIn, Amazon, and Snap.
Meta was my first industry role. I joined directly from academia as an IC6 Research Data Scientist (Applied Scientist) in the infrustructure org. My work was quite interesting: I was building ML algorithms for ranking Android code before it was shipped (as it happens, the order of the shipped code affects latency). Transitioning from academia was quite challenging, and I had to learn a lot, but overall I did quite well as a Rsearch Data Scientist.
When Meta had its "moment" (second layoffs) in April 2023, my original research data scientist group got obliterated. Most of research data scientists were laid off, but some of us were given the option to transition to product data science roles. I took this option and moved to monetization. This was my first experience as a product data scientist, and I hated it from the very begining. Instead of modeling and problem solving, PDS focuses on metrics, strategy, communication. It is fine if this is what you like, but it wasn't for me. I immediately started looking for a new role as an applied scientist or a machine learning engineer. Transitioning internally to MLE was really not an option (I have seen people getting fired because they decided to change teams) plus my interaction with MLEs in monetization at Meta pretty much persuaded me to look elsewhere (more engineering than modeling).
It took me about 9 months where I was interviewing on the side (very, very challenging thing to do) to find a role I was really excited about. At that time I had 4 offers and 8+ final scheduled loops that I chose to cancel. One offer was from Google, but it was for a DS role so I decided that I really wanted to put the right title (i.e., Applied Scientist) on my resume. In addition, the role I actually accepted was an almost ideal match with my academic research, so I was really excited to get back to work I was trully passionate about.
I am extremely happy with my current role: in fact I get to do things that I would do as a hobby -- of course there is stress and pressure to deliver impact, but it is a lot more fun that what I was doing at Meta.
My 2 cents: Meta is fine if you like the role and you are right out of school (new grad) without a family. It is quite easy to grow and learn a lot, and quite stress free if you are willing to put 50+ hours. It also pays extremely well. It becomes challenging when you join at higher levels, especially if there is a role mismatch (like in my case with product ds) and you have a family (which means you can't work more than 40 hours a week).
I do this for fun, which means I don't work on it every day. My job is quite intense, and I have two young kids. At most, I find 4-5 hours a week to work on this. A challenging solution can take several hours to write up, so I end up adding a few new questions every week.
The same goes for updating the front end or adding functionality on the platform.
I am confident, that by the end of 2025 the platform will be in great shape.
A lot. To get some ideas check on levels.
Applied scientists and MLEs tend to make more than DS and PDS. In most cases, they make the same or more than SWEs. DS and PDS tend to make a little less. At Meta, Research Data Scientists were making 85%-90%% of SWE, whereas Product data scientists were making 65% of SWEs. In my current company, applied scientists are part of the Eng org, and make the same as SWEs.
Absolutely. If you have seen an interview quwestion you are not certain about its solution (or you don't trust chatgpt) feel free to email me at comments@madinterview.com
Have you used chatgpt (or any other LLM) to build anything with a decent level of complexity? Don't buy the hype. We are far, very far, from having unsupervised agents writing trusted software. Check out this website for some serious, unhyped, AI talk.
It is true however that as a DS (and more so as an AS/MLE) our job has changed significantly. I do a fair amount of prompting in my current role, which wouldn't have been the case 2 years ago.
For product DS/DS roles you must know everything in Kohavi's trustworthy online controlled experiments book.
For probability and statisctis, a great book is Pishro-Nik's Introduction to Probability, Statistics, and Random Processes.
For ML fundamentals, a great book book is Murphy's probabilistic machine learning.
For deep learning, I really liked Bishop's new book.
For ML systems design, Chip's Designing Machine Learning Systems: An Iterative Process for Production-Ready Applications is an amazing book.
For understanding LLMs, there is only one book you should read: Build a Large Language Model (From Scratch) by Sebastian Raschka.