LongxingTan/Machine-learning-interview: A Chinese-Language Interview Guide Split by Job Title

The problem is track selection, not missing chapters

Most interview repositories dump everything into one flat list. A candidate for a data engineer role then wades through ML system design, and an MLE candidate spends a week on product sense. This repository does the opposite. The README presents four audience tables, and each table names a different sequence of chapters. The MLE and data scientist (model) track is Chapter 1 Data Structures and Algorithms, Chapter 2 ML Breadth / Depth, Chapter 3 ML System Design, Chapter 4 Behavioral Question. The software engineer track swaps ML system design for System Design and adds Object Oriented Design as a fourth chapter. The data scientist (analytics) track drops algorithms entirely in favor of Statistical and Causal Inference, ML Breadth / Depth, Product Sense, and SQL. The data engineer track keeps algorithms and SQL, adds OOD, and ends with behavioral questions. The audience is stated explicitly in the README: software engineer (SDE), machine learning engineer (MLE), data scientist (DS), applied scientist (AS), AI engineer, and data engineer. The README also frames the interview itself with a short line about communication, quoted as logic, framework thinking, and highlighting key points. That framing is the closest thing to a stated philosophy in the material.

What the repository actually contains versus what it links to

The README is a routing table, not a textbook. Every chapter entry is a relative markdown link into the repo tree: 01_leetcode/README.md, 02_ml/README.md, 03_system/03_ml/README.md, 04_bq/README.md, 03_system/README.md, 03_system/01_ood/README.md, 02_ml/00_ml_math.md, 02_ml/97_product_sense.md, and 01_leetcode/99_SQL/README.md. Those paths tell you the top-level layout: a leetcode directory with a nested SQL subdirectory, an ml directory holding both a math file and a product sense file at numbered prefixes, a system directory with an OOD subdirectory and a separate ML system design subdirectory, and a bq directory for behavioral questions. The README also points to a free book at longxingtan.gitbook.io/ml-interview, described as free. So the repository appears to function as the source or companion for a GitBook rather than as a standalone reading experience. The material does not say whether the GitBook is generated from these files or maintained separately, and it does not state how complete any given chapter is. Treat the directory names as a promise about topics, not as evidence of depth.

There is no install step, and that is the point

Nothing in the supplied material describes a build, a package, a runtime, or a configuration file. There is no package.json, no requirements.txt, no Makefile, and no config keys mentioned. The primary language field is unknown and no releases were retrieved. The way you get this running is: clone the repository, then read markdown. The README gives the entry point as a GitBook URL rather than a local command, so the fastest path is to open the GitBook in a browser and use the repo only when you want the raw files. If you do clone, the four track tables in the README are your reading order, and the relative links above are the files to open. That is the whole onboarding. Anyone expecting an interactive problem set, a test harness, or a graded exercise runner will not find one described here. The repository is a document collection with a navigation layer, and the navigation layer is the part that was actually designed.

The four-track split has a cost: duplicated and uneven depth

Look at the tables side by side and the overlap is heavy. Data Structures and Algorithms appears in three of the four tracks. ML Breadth / Depth appears in two. SQL appears in two. Behavioral Question appears in three. Object Oriented Design appears in two. Only the analytics data scientist track avoids algorithms, and only the MLE track includes ML System Design. This means a reader who is undecided between MLE and SDE is reading largely the same Chapter 1 either way. The split is useful for orientation and wasteful for anyone who wants to study once. The second cost is depth. The README gives a single markdown file for statistical and causal inference (00_ml_math.md) and a single file for product sense (97_product_sense.md). Causal inference and product sense are both large interview areas, and one file each is a thin container for either. The numbered prefixes suggest the author intends to add files around them, but the material as given does not show that the surrounding files exist. The third cost is language. The README is in Chinese and the description states it is a Chinese interview guide aimed at overseas and foreign-company CS roles. If your interview will be conducted in English, this repository prepares you in the wrong language for the part of the interview that is actually assessed on communication.

Overlap with general-purpose interview repositories, and where this one diverges

The obvious alternative for the algorithms chapter is a general-purpose problem archive such as LeetCode itself, or a broad English-language preparation repository. The difference in approach is real. A problem archive gives you a large pool of exercises with an online judge and no opinion about which ones matter for your title. This repository gives you no judge and no exercises of its own in the algorithms chapter; it gives you a reading order keyed to a job title. That is a curation claim, not a content claim. The second alternative is a full textbook on ML system design or causal inference. A textbook goes deep on one area; this repository's value is that it tells you which areas to visit and in what order, then hands you off. If you already know your target role and your weak areas, the routing table adds little and you should go straight to the source material. If you are early in preparation and genuinely unsure whether you are an MLE candidate or a data engineer candidate, the four tables are the most useful thing here, because they make the distinction concrete through chapter lists rather than through job descriptions.

Maintenance, licensing, and what the record does not show

The repository is not archived and the last push is dated 2026-05-30, so it is being touched. No releases were retrieved, which is expected for a documentation repository but means there is no versioned snapshot to pin your study plan to. There is no stated license in the material provided. The README calls the GitBook free, but free to read is not the same as licensed for reuse, and the absence of a license file matters if you intend to fork the content, translate it, or fold it into internal training material. The material does not state a license, so verify that directly on the default branch before reusing anything beyond personal reading. Maintenance cost for a user is close to zero: there is nothing to upgrade, no dependency to track, and no API to break. The cost is on the other side. A repository of markdown links into a GitBook can drift, and if the GitBook is edited independently, the relative links in the README may point at files that no longer match what readers see online. The material does not tell us whether that drift has occurred.

Who this fits and what to confirm before committing study time

Use it if you are targeting a Chinese-language interview loop and you want one page that maps your title to a chapter sequence. The four tables are the deliverable, and they are specific enough to act on. Do not use it as your only preparation source if your loop is in English, if you need graded practice problems, or if you need a self-contained corpus that works offline without the GitBook. Before you commit weeks to it, do three checks. Open longxingtan.gitbook.io/ml-interview and confirm the chapters named in your track are actually written out rather than stubbed. Open 02_ml/00_ml_math.md and 02_ml/97_product_sense.md and judge whether one file each covers the statistical and causal inference and product sense depth your target company asks for. Then look for a LICENSE file on the main branch, because the README does not name one and that determines whether you can do anything with the text other than read it.

Editorial conclusion

Adopt this if you are preparing for a Chinese-language interview loop at an overseas or foreign-owned company and want a single index that tells you which chapters apply to your target title. Skip it if you need English-language material, a self-contained offline corpus, or a permissively licensed text you can reuse. Before relying on it, open the GitBook at longxingtan.gitbook.io/ml-interview and confirm the chapters are populated, then check whether a LICENSE file exists on the main branch, because the README does not state one.

Official sources

1. Issues
2. LongxingTan/Machine-learning-interview on GitHub
3. Project website
4. README