Module Guide & Role Landscape

Remember this one-liner from the page: the career module solves two things—first, see what roles the market has and what each requires; then, prepare yourself against the gaps. This article is the navigation map for those two things: the role landscape gives you market intelligence, the decision checklist helps you position yourself, and the content map tells you which four pages to go deep on.

The most dangerous sequence error in job searching is "effort first, positioning later": grinding model projects for three months only to find your target role doesn't test any of them. The career module pulls you back on track—do information work first, then study work.

1. Module Positioning: See the Market, Then Prepare Yourself

The four articles in this module form a job search pipeline, where each stop's output is the next stop's input:

Your situation: want to enter / switch / jump into ML
          │
          ▼
┌─────────────────────────────────────────────┐
│ ① See the market ── JD List                 │
│    What roles are hiring? What requirements  │
│    keep appearing in JDs?                    │
│    Salary bands, skill weight, degree bar    │
└─────────────────────────────────────────────┘
          │
          ▼
┌─────────────────────────────────────────────┐
│ ② See yourself ── Skill Map                 │
│    Map JD skill words to "know / don't know  │
│    / partially know", generate a personal     │
│    study gap list instead of reading from    │
│    scratch                                   │
└─────────────────────────────────────────────┘
          │
          ▼
┌─────────────────────────────────────────────┐
│ ③ Package yourself ── Resume Analysis       │
│    Rewrite project descriptions from the     │
│    interviewer's perspective: highlight      │
│    model selection, business results,        │
│    failures and retrospectives               │
└─────────────────────────────────────────────┘
          │
          ▼
┌─────────────────────────────────────────────┐
│ ④ Prove yourself ── Interview Questions     │
│    High-frequency questions + answer         │
│    frameworks; self-test last, not day one   │
└─────────────────────────────────────────────┘

This sequence isn't invented by this site; it's the main thread design of the Job Search Sprint Plan: start from the end, reverse-engineer learning content from JDs and interview questions—not aiming for comprehensiveness, only that all high-frequency check points are covered. If you only have two to three weeks, follow this main thread; if time is ample, cross this main thread with the site's regular path from "guide → core knowledge → case studies → practice."

Relationship with other modules

The career module does not repeat knowledge. It does two things: market intelligence + job search methodology. For conceptual content, go to Core Knowledge to clarify boundaries, or the Glossary for definitions. The most common mistake in the job search phase is spending time on "one more chapter" instead of "organizing what you already know"—the career module is designed for the latter.

2. Role Landscape: Full Panorama of Nine Role Types

Start with the summary table, then break down each one.

Role	Common English Name	One-line positioning	Main output
Algorithm Engineer	Algorithm Engineer	A uniquely domestic broad term: responsible from modeling to deployment	Models that can be deployed and iterated
ML Engineer	ML Engineer / MLE	Engineering expert turning models into reliable systems	Stable, efficient training/inference pipelines
Data Scientist	Data Scientist	Statistical modeler using data to answer business questions	Credible conclusions and decision recommendations
Data Analyst	Data Analyst	The gatekeeper translating data into business language	Dashboards, metrics, and insights
LLM Engineer	LLM Engineer	The fastest-growing new role post-2022	Product capability built on large models
DL Researcher	DL Researcher	The scientist role inventing new methods	Papers, patents, new algorithms
NLP Engineer	NLP Engineer	The modeler handling natural language	Text understanding and generation systems
CV Engineer	CV Engineer	The modeler enabling machines to "see" images	Detection/recognition/generation systems
Recommender Engineer	Recommender Engineer	The modeler optimizing ranking and distribution	Higher-conversion recommendation systems

How to read salary numbers (please read this first)

Salary figures in this section are coarse-grained synthesized ranges based on publicly available job platform data, third-party statistics from levels.fyi, etc., at the writing date (dataAsOf: 2026-08), for reference only when choosing a direction, and do not constitute precise quotes. "Tier-1 domestic cities" refers to full-time annual salary in RMB (covering Beijing, Shanghai, Shenzhen, Hangzhou, etc.); "overseas" is primarily U.S. tech companies (in USD). Same-role salaries can vary by several times due to degree, experience, company, and interview performance; beyond a quarter, treat as trend reference only, and use actual offers and latest statistics for precise numbers.

1. Algorithm Engineer —— The #1 Keyword in Domestic Hiring

"Algorithm Engineer" is a uniquely domestic role name with a very broad scope, often the first stop for fresh grads and career-switchers. Responsibilities span the full chain from reading requirements → defining the problem → feature engineering → training → evaluation → deployment → monitoring and iteration. Most companies' algorithm engineers lean toward business modeling, using both tree models and deep models, and LLM capabilities (Prompt, RAG) are increasingly appearing in JDs.

• Core skills: Python and DL frameworks (PyTorch as mainstream), ML fundamentals (supervised/unsupervised/evaluation), data structures and algorithms (for written tests), feature engineering, business understanding.
• Typical salary: Tier-1 domestic cities ~300K–700K RMB/year (higher at big company core departments, significantly lower at small companies and outsourcing); overseas similar roles are ML Engineer or Applied Scientist, ~$150K–$250K/year.
• Pages on this site: The knowledge foundation is the site's core knowledge and case study chapters; for job prep, cross-reference the JD List directly and use the Skill Map to generate a study gap list.

2. ML Engineer —— Upstream of Models, Downstream of Engineering

ML Engineer and Algorithm Engineer are often conflated, but in mature teams the division differs: algorithm engineers lean toward "building more accurate models," ML engineers lean toward "making models run stably and reliably in production." Responsibilities include feature pipelines, training job scheduling, model serving, monitoring and alerting, A/B testing platforms, and increasingly heavy MLOps (data versioning, model versioning, CI/CD, drift detection).

• Core skills: Python, SQL, data engineering (Spark, Flink, etc., depending on team), MLOps toolchain, containers and cloud services, distributed training, systems engineering mindset.
• Typical salary: Tier-1 domestic cities ~350K–600K RMB/year; overseas ~$150K–$250K/year (MLE at U.S. big companies is one of the fastest-growing salary directions).
• Pages on this site: Modeling parts look at core knowledge; engineering practice is the backbone of the practice chapter; interview focus is "can you explain the full pipeline clearly," self-test with Interview Questions.

3. Data Scientist —— Using Data to Answer "What to do"

A data scientist's core isn't modeling itself, but using data to make decisions: defining problems, designing metrics, running experiments (A/B), using statistical inference and modeling to answer business questions, and communicating conclusions to business stakeholders. Compared to algorithm engineers, this role is closer to business and causality, with significantly higher weight on statistics and communication skills.

• Core skills: Statistics and hypothesis testing, experimental design, SQL, Python/R, ML modeling, causal inference, data visualization and business communication.
• Typical salary: Tier-1 domestic cities ~250K–500K RMB/year; overseas ~$130K–$220K/year.
• Pages on this site: Statistical and model concepts in the core knowledge chapter, unified terminology in the Glossary; this role's "role-fit" differs significantly from algorithm roles—read Resume Analysis before applying to confirm you're writing a data science story rather than an algorithm story.

4. Data Analyst —— The Translator from Data to Business

Data analysts are the easiest to enter and also the most underestimated: querying, cleaning, dashboards, metric definitions, attribution analysis, visualization, and some teams also handle event tracking design and experiment support. They almost never write models, but their requirement for business understanding and SQL may be the hardest of all roles.

• Core skills: SQL (hard threshold), Excel/BI tools (Tableau, Power BI, etc.), Python (pandas, visualization), statistical foundations, business metrics and funnel thinking.
• Typical salary: Tier-1 domestic cities ~150K–350K RMB/year; overseas ~$80K–$150K/year.
• Pages on this site: This site doesn't teach data analysis tools specifically, but this role is often a stepping stone into ML; concepts and terminology in the Glossary suffice; trade-offs in the switching path are on this page, Section 4.

5. LLM Engineer —— The Fastest Growth Post-2022

The LLM engineer role only became large-scale after ChatGPT in late 2022. Its core is integrating general-purpose large models (GPT, Claude, open-source models like Qwen, Llama, etc.) into products: Prompt engineering, RAG, Agent/workflows, fine-tuning (LoRA, etc.), evaluation set construction, cost and latency optimization. It doesn't require you to train large models from scratch, but it does require you to know model capability boundaries and combine them with engineering skills.

• Core skills: Python, LLM API and open-source model deployment, RAG and vector search, evaluation frameworks (the most standout and most overlooked part), Agent frameworks, productization and engineering skills.
• Typical salary: Tier-1 domestic cities ~400K–1M RMB/year (senior and scarce talent carry high premiums); overseas ~$180K–$300K/year.
• Pages on this site: The foundation is still your existing ML and DL background; "LLM application" mapping to site pages is findable in the Skill Map.

6. DL Researcher —— The Role Closest to Papers

Researchers' goal is producing new methods: reproducing papers, proposing improvements, running large-scale experiments, publishing papers or patents. Most common at big company research institutes, autonomous driving / multimodal labs, and top-tier startups. Demands on math and paper-reading are the highest among the nine roles; demands on engineering and business are the lowest.

• Core skills: Deep learning principles and latest papers, solid math (linear algebra, probability, optimization), experimental design and ablation analysis, PyTorch/JAX, English paper reading and writing.
• Typical salary: Tier-1 domestic cities ~400K–900K RMB/year; overseas ~$150K–$280K/year.
• Pages on this site: The site's paper chapters and DL-related content are the entry foundation; but the hiring bar for researcher roles (top-tier papers, competition rankings) far exceeds anything an article can cover—base your target on the hard criteria in the JD List.

7. NLP Engineer —— The Builder of Language Intelligence

NLP engineers do text understanding and generation: text classification, NER, semantic search, machine translation, dialogue systems, and an increasingly large share of LLM applications. This role's special trait is heavy overlap with LLM engineers—in recent years, many NLP roles are "classic NLP + LLM" hybrids.

• Core skills: Python, classic NLP (tokenization, word embeddings, sequence labeling), Transformer principles, LLM applications, data and evaluation; for Chinese scenarios, also understanding of tokenization and corpus characteristics.
• Typical salary: Tier-1 domestic cities ~300K–600K RMB/year; overseas ~$150K–$250K/year.
• Pages on this site: DL and Transformer concepts are the foundation; terminology in the Glossary.

8. CV Engineer —— Enabling Machines to See the World

CV engineers do perception and generation for images/video: object detection, image segmentation, face recognition, video understanding, image generation. The tech stack is concentrated and iterates fast (CNN → Transformer → multimodal), with mature open-source frameworks (MMDetection, Ultralytics, etc.), and the entry barrier is relatively friendly among the nine roles.

• Core skills: Python, DL frameworks, CNN/Transformer principles, detection/segmentation/tracking paradigms, multimodal and generative models, data annotation and evaluation.
• Typical salary: Tier-1 domestic cities ~300K–600K RMB/year; overseas ~$150K–$250K/year.
• Pages on this site: Image modeling mechanisms in the case study chapter, quick foundation-building with the Glossary.

9. Recommender Engineer —— The Modeling Role Closest to Business Results

Recommender engineers do ranking and distribution: recall, pre-ranking, ranking, re-ranking, feature engineering, online learning, A/B testing. Their uniqueness is that performance is directly settled in business metrics (CTR, conversion rate, session duration), so communication, experimentation, and engineering skills weigh more heavily than in general modeling roles.

• Core skills: Python, CTR prediction model evolution (LR → GBDT → DNN → Transformer—must be able to explain), Embeddings and two-tower, feature engineering, experimentation and causal thinking, big data processing.
• Typical salary: Tier-1 domestic cities ~350K–700K RMB/year; overseas ~$160K–$260K/year.
• Pages on this site: Supervised learning and evaluation are the foundation; the rigor of experimental design can be cross-referenced against Interview Questions for questions like "how to prove the improvement you brought is real."

3. Role Skill Radar: Modeling · Engineering · Business · Math

Put the nine roles into a four-dimension coordinate system, and their personality differences become visible at a glance. The four dimensions are defined as:

             Modeling (model selection / tuning / performance optimization)
                       ▲
                      /|\
                      │
      Math ──────────┼────────── Engineering
  (statistics / linear algebra /      (code quality / deployment /
   probability / derivation)          MLOps / pipelines)
                      │
                     \|/
             Business (metrics / communication / experimentation / business sense)

Each axis weight uses 1–5 (1 = basically not needed, 5 = core capability of the role):

Role	Modeling	Engineering	Business	Math	Role Personality
Algorithm Engineer	4	4	2	3	Generalist
ML Engineer	3	5	2	2	Engineer
Data Scientist	3	2	4	4	Business + Stats
Data Analyst	1	2	4	3	Business
LLM Engineer	3	4	3	2	Engineering + Product
DL Researcher	5	2	1	4	Scientist
NLP Engineer	4	3	2	3	Builder
CV Engineer	4	3	2	3	Builder
Recommender Engineer	4	3	3	3	Business + Builder

How to use this table

Don't just stare at "which row has the highest numbers." Look at which column you can focus on long-term: modeling-strong people tolerate the boredom of tuning; engineering-strong people enjoy the stability of systems running smoothly; business-strong people need the sense of achievement from standing in the business stakeholder's shoes; math-strong people enjoy the certainty of derivation. Rank "which kind of work I'm more willing to do long-term" above "which role is trendier."

4. Which Direction Should You Choose: Decision Checklist

No standard answer, but there are standard questions. Go through the three checklists below, and your checked answers will converge on one or two roles.

1. By Background: Where Are You Now?

• [ ] CS / software background, coding for years, want to transition into ML → prioritize ML Engineer / Algorithm Engineer
• [ ] Math / stats / physics background, comfortable with derivation but less engineering → prioritize Data Scientist / DL Researcher
• [ ] Business / commerce / operations background, SQL proficient → prioritize Data Analyst, then transition toward Data Scientist based on interest
• [ ] Fresh grad with no industry experience, want to get interviews in the shortest time → Algorithm Engineer has the most JD openings and is the most universal entry point domestically
• [ ] Already in traditional software, want to ride the LLM wave → LLM Engineer (existing engineering experience is a significant plus)

2. By Interest: What Kind of "completion feeling" do you enjoy?

• [ ] "Model metrics went up another 0.3" → modeling inclination: algorithm / NLP / CV / recommender all work
• [ ] "This pipeline ran steadily for three months without alerting" → engineering inclination: ML Engineer
• [ ] "I helped the business side clarify the problem definition" → business inclination: Data Scientist / Data Analyst
• [ ] "I took a paper from reproduction to improvement" → research inclination: DL Researcher
• [ ] "I assembled a new feature using off-the-shelf large models" → product + tech inclination: LLM Engineer

3. By Trend: How much certainty are you willing to pay for?

The market structure for ML roles is changing. Several facts worth including in your decision (as of dataAsOf: 2026-08):

• Pure "tuning-type" general algorithm roles are decreasing; "engineering + LLM application" hybrid demand is rising, with the proportion of LLM keywords in domestic and overseas JDs climbing rapidly;
• Data science/analysis roles are continuously elevating demands for business and causality; purely tool-based analysis is being squeezed by BI tools;
• Researcher role barriers are rising, with top-tier conference papers shifting from nice-to-have to hard requirement;
• Vertical directions like recommendation / CV / NLP haven't disappeared, but they're all adding multimodal and LLM capabilities.

Look at the results from all three sets together: if all three point in the same direction, go for it decisively. If directions conflict, use 「background as the floor, interest for distance, trend as the weight」 to set priorities—background determines whether you can walk through the door, interest determines how far you can go, and the trend only determines how crowded the door is.

Don't treat "role names" as identity

The same name can mean completely different work at different companies: Company A's "Algorithm Engineer" might write SQL all day, while Company B's "Data Scientist" might tune models all day. Reading JDs is always more important than reading role names—which is why the first page of this module is the JD List, not a "role encyclopedia."

5. Content Map for This Module: What Each of the Four Pages Does

The career module has four articles, corresponding to four actions in the job search process:

Page	One-line positioning	When to use
JD List	Consolidates real JDs at major domestic and overseas companies, breaks down high-frequency requirements and thresholds	Day 1 of job search: first understand what the market wants
Skill Map	Maps JD skill words to pages on this site, generates a personal study gap list	After sizing up the market: position your own gaps
Resume Analysis	Resume review criteria from the interviewer's perspective, with project description rewrite examples	When preparing applications: turn experiences into evidence
Interview Questions	High-frequency Q&A + answer frameworks, self-test	One week before interviews: final self-test

The usage order is the main thread, not an option

The Job Search Sprint Plan has already ordered the four pages into a two-week main thread: JD List → Skill Map → Resume Analysis → Interview Questions. The output of each page is the input for the next. Reading out of order loses half the value.

6. Two Reminders

1. Check dataAsOf Before Trusting Numbers

Roles, salaries, JD requirements, and technology keywords all have shelf lives in this module: ML has a very short half-life—any role's JD might be rewritten twice in a year, and salary bands fluctuate with the market. Therefore, all time-sensitive pages in this module are marked with dataAsOf (data as-of month) in the frontmatter, and the page header also displays it. Before citing specific numbers from here, check that date:

• Within a quarter of the current date → can be used as current reference;
• Beyond a quarter → treat as "trend reference," and verify with official sources;
• Conceptual content (bias-variance, overfitting) decays very slowly and is not subject to this constraint.

2. JD Description vs. Actual Work: Two Versions of the World

A JD describes "what they wish you'd be"; actual daily work is "what you're actually needed for." There's a systematic gap:

• JD says "proficient in deep learning," the reality is mostly data cleaning and feature tweaking—model tuning is only a small part of the job;
• JD says "optimize accuracy," the real evaluation is business metrics—2 points of AUC increase but no conversion improvement, and you still don't pass review;
• JD says "independently owns," the reality is you're one piece in a massive system—it's far more credible to explain "the part I owned" than to claim "I own the entire system";
• JD says "LLM experience preferred," the reality is the team itself hasn't solidified—honestly discussing your understanding of this uncertainty is itself a plus.

The only way to cope: do intelligence on the target team before interviews (interview experiences, conversations with current employees, product experience). Treat JDs as hypotheses and verify with research. There's a checklist of "questions to ask the interviewer" in Interview Questions—that's exactly what it's for.

7. Further Reading

Continue on this site

• Job Search Sprint Plan —— The complete two-week preparation main thread that the career module lives within
• ML vs AI vs Deep Learning vs Data Science —— Clarify the conceptual boundaries behind role names
• Glossary —— Check definitions when reading JDs and interview questions

Reference materials (real sources, for verifying salary and trends)

• levels.fyi —— U.S. tech company salary database, query by company/level/city, reliable source for overseas salary
• Stack Overflow Developer Survey —— Annual developer survey, including ML technology usage rates and salary distribution
• U.S. Bureau of Labor Statistics: Data Scientist Occupational Outlook —— Official growth projections and salary statistics for data scientist roles in the U.S.
• Kaggle State of Data Science & Machine Learning —— Annual global ML practitioner survey, for role distribution and tech ecosystem changes
• Zhaopin China Job Market Prosperity Report (CIER) —— Domestic job market prosperity and industry demand, Chinese-perspective reference