This Post Has 10 Comments

1. Comprehensive resource for understanding ML systems
   I recently purchased this book with the intention of gaining a deeper understanding of how ML systems are built in practice. I was pleased with what I found in this book.The book consists of 11 chapters, starting with an introduction that outlines a framework for approaching ML system design interview questions. The following 10 chapters each delve into a real-world system that is commonly used in the industry.Pros: – Practical Focus: The book’s main strength lies in its focus on practical examples, which helps readers to better understand the concepts and apply them in real-world situations. This approach is particularly useful for preparing for ML system design interviews, where resources on this topic can be limited. – Clear Explanations: Each chapter is well-explained, with clear examples and case studies that effectively illustrate the concepts. The book covers a broad range of topics, from modeling algorithms to data pipelines and practical tips for scaling ML systems. The authors have done an excellent job of discussing different solutions and the trade-offs involved in building ML systems. – Interview-oriented: The authors provide practical tips and guidance on how to approach machine learning system design interview questions and what to expect during the interview process. – Easy to Navigate: The book is well-organized and easy to navigate, with clear headings and subheadings that make it easy to find the information you need. The writing style is clear and concise, and the authors do an excellent job of explaining complex concepts in a simple and understandable way.Cons: – Limited ML Fundamentals Coverage: The book does not cover ML fundamentals and is not suitable for those who want to learn the basics of ML and related concepts. – Domain Specificity: The authors could have covered more examples from different domains, as there are several important systems that are not covered in the book, such as generative AI, language modeling, and ETA systems. – The book does not delve deeply into complex topics, making it potentially less suitable for staff-level engineers and above.Overall, I found this book to be a comprehensive resource for preparing for technical ML interviews and for gaining a high-level understanding of ML systems. I highly recommend it.

   Reply

2. Excellent reference for bridging theory and practice
   Really excellent breakdown of a number of case studies. Good reference for working ML engineers as well as students who intend to enter industry.The heuristics are well explained. In particular, the end of chapter diagrams are helpful for seeing common threads between the examples. There are many insights about how to think practically about training dataset construction and serving pipelines that you will not find in ML textbooks.Where I would make improvements:- the graph neural network example (PYMK) should be more fleshed out when it comes to how the architecture actually works. Unlike the other examples, I had to look at the reference articles in this section pretty frequently, to the point that at least some of their info should have been included- I would like to have seen some more explicit formulas in certain places for people who best understand functions by reading them directly. Off the top of my head, the discussion of focal loss, and various offline eval metrics, would benefit from adding these.- Sometimes when multiple dependent models were required, the discussion of how they linked up could be expanded, e.g. the regression + NN design for IDing license plates in street view

   Reply

3. Really really good book. Thank you Alex and Ali!
   Absolutely worth the money!

   Reply

4. Great tool for T/PMs and early-mid career engineers
   This book makes a valiant attempt at describing software architectures holistically, but doesn’t really add more value than what can already be found online. I was hoping for some additional language on how to manage the conversation itself for each example, as driving the convo is nearly 50% of the skill set required for a good interview. The book gives an example convo during the requirements gathering by step for each example, but doesn’t revisit additional questions or gotchas later.It also doesn’t talk much at all about how the ml system fits in with the overall system design, which is a different tactic that could have made this book more interesting than the current material online.That being said, this book helped me get where I needed to go, and for that reason, I give it four stars. I say that as a TPM (and former lead engineer), where the expectations for going into technical details are not quite as high as a senior or staff level engineer. As such, I only fully recommend this book for early- mid career engineers, and TPMs and pms.Unless you haven’t interviewed in a 5+ years, senior ml engineers should have the expectation that this is a mere starting point. If you interview others often, you likely won’t need this book at all and should instead search for deeper technical details and trade off considerations elsewhere.

   Reply

5. Great case studies, not just for interview prep
   The book has 11 chapters. The first chapter presents the fundamentals, and the remaining covers ten use cases. The patterns I’ve learned have helped me think more critically. I highly recommend it.Good:It is a great resource for communicating decisions in a way that is well-organized and universally understood. Two features I really liked:1) Mind maps for each design2) Offering a dependable and repeatable framework for tackling different ML systems. Having a strong framework is crucial, allowing the practitioner to focus on the unique aspects of the system.Bad:My wish was that the book could cover more aspects of the ML interview, such as ML coding and ML theory.Other resources:It is a tough job market out there. My friends and I have been preparing for job interviews for three months. Below is the list of materials we found helpful. Good luck, everyone!- Stanford CS229: Machine Learning- Deep Learning book- Designing machine learning systems book by Chip Huyen- She also maintains a great GitHub repo- Made with ML- ML system design interview guide by Patrick Halina- Industry papers. Tiktok, YouTube, and Instagram all released great papers about recommendation systems.

   Reply

6. This book really helped for preparing for my interview at a big tech company. Would 100% recommend.

   Reply

7. Great book.The authors began by writing an extensive overview of machine learning systems from theoretical clarification of requirements to advanced monitoring and infrastructure. They built on that and introduced several examples of machine learning system design questions you could encounter such as recommender systems, ad click prediction, search problems, etc.Overall, I highly recommend this book

   Reply

8. Book tries to give an overview of many different systems that use ML, but to my taste, lacks proper structure within topics to certain degree, diversity in topics (k-nearest neighbors is repeated many times), deep dive (it just mentions important issues many times (e.g. bias), but never tries to explain a good approach to solve them) and etc. Gives you the impression that authors were in a hurry to publish the book. Overall not bad and good starting point for junior ML engineers.

   Reply

9. I haven’t read any ML books as bad as it is. So many low-level mistakes were made in this book. Clearly, the author doesn’t have systematic knowledge about machine learning/deep learning. It looks like this book was written by multiple useless managers who don’t understand ML.

   Reply

10. This is a great book to read. Book is clearly printed and delivered quickly and perfectly. Recommend to purchase.

    Reply