Theme
JD List: Open Roles at Major Domestic and Overseas Companies
In one sentence: JDs (Job Descriptions) are raw text of market demand. This page selects 8 representative roles and lays out "key responsibilities / hard requirements / bonus qualifications" for you, along with skill keyword frequency statistics, so you can see at a glance what the market keeps asking for.
Timeliness note: Role information and salaries on this page are representative samples from mid-2026, data current as of 2026-08. Verify before citing. Specific roles are subject to the company's official recruitment site on the day.
Read the Module Overview & Job Landscape first for role classification, then go through the breakdowns below.
1. How to Read a JD: Three-Layer Method
For any JD, break it down in three layers:
- Responsibility layer (What): What do you do daily after joining — modeling, deployment, tuning, or research?
- Threshold layer (Must): Hard requirements — degree, years of experience, must-have skills. Self-assess against this layer before applying;
- Differentiation layer (Plus): Bonus qualifications — what determines whether you get pulled out of the pile of candidates.
Write the three layers into three columns and you have a "role skill keyword list," which feeds into the next page, JD Knowledge Map. Each role below follows this format.
2. Domestic Big-Tech Role Breakdowns
Role 1: LLM Algorithm Engineer (Application Direction)
Typical hiring companies: Top internet companies' LLM teams, AI unicorns.
| Layer | Content |
|---|---|
| Key responsibilities | LLM deployment in business scenarios (dialogue, content generation, agents); instruction fine-tuning and RLHF alignment; RAG retrieval enhancement and evaluation; online effect analysis and iteration |
| Hard requirements | Master's degree or above; solid Transformer and training theory foundation; proficient in PyTorch and distributed training; complete LLM project experience |
| Bonus | Top-tier paper publications; RLHF deployment experience; inference acceleration (vLLM / quantization); agent tool-calling engineering experience |
Linked pages: Large Language Models (LLMs) + Training Recipes & Hyperparameter Tuning.
Role 2: Computer Vision (CV) Algorithm Engineer
Typical hiring companies: E-commerce, autonomous driving, security vendors.
| Layer | Content |
|---|---|
| Key responsibilities | Model selection and tuning for object detection / segmentation / tracking / OCR; data annotation workflow design; model distillation and edge-device deployment; online metric monitoring |
| Hard requirements | Familiarity with CNNs and common detection/segmentation frameworks; PyTorch proficiency; data augmentation and tuning experience; ability to independently complete the full pipeline from data to evaluation |
| Bonus | Edge-device inference optimization experience; open-source contributions; competition rankings; self-supervised / large-scale pre-training experience |
Linked pages: CNNs & Computer Vision + Datasets & Tool Archive.
Role 3: NLP Algorithm Engineer
Typical hiring companies: Search, recommendation, customer service, and financial text analysis teams.
| Layer | Content |
|---|---|
| Key responsibilities | Text classification / extraction / retrieval / generation; LLM fine-tuning and prompt optimization; NLP data pipelines; effect evaluation systems |
| Hard requirements | Proficiency in Transformers and their variants; text data processing skills; PyTorch / HuggingFace ecosystem; understanding of evaluation metrics (precision/recall/F1, NDCG) |
| Bonus | Retrieval-augmented generation (RAG) experience; multilingual model experience; large-scale pre-training participation |
Linked pages: Transformer Architecture + Deep Learning Evaluation & Experimentation.
Role 4: Recommendation Algorithm Engineer
Typical hiring companies: Content platforms, e-commerce, local-life services.
| Layer | Content |
|---|---|
| Key responsibilities | Full-link modeling for recall / coarse ranking / fine ranking; user and item representation learning; feature engineering and online learning; A/B experiment analysis |
| Hard requirements | Familiarity with CTR prediction models (DeepFM, etc.); large-scale sparse feature handling; PyTorch and big-data tools (Spark/Hive); deployment capability |
| Bonus | Reinforcement learning in recommendation; graph neural network-based recall; multi-objective optimization experience |
Linked pages: Deep Learning Recommender Systems + Graph Neural Networks.
Role 5: Algorithm Engineer (Platform / General)
Typical hiring companies: Most big-tech general roles, JDs are typically broad.
| Layer | Content |
|---|---|
| Key responsibilities | Modeling needs across multiple business lines; general model capability building; platformizing training/evaluation workflows |
| Hard requirements | Solid deep learning theory foundation; PyTorch proficiency; SQL and big-data basics; 3+ years relevant experience |
| Bonus | Multimodal experience; MLOps platform building experience; cross-team collaboration and project leadership |
Linked pages: Deep Learning System Anatomy + MLOps & Model Deployment.
3. Overseas Role Breakdowns
Overseas JDs have more specific naming. First distinguish three types: MLE (Machine Learning Engineer, engineering-leaning), Research Scientist (research/-paper-leaning), and Applied Scientist (balance of research and deployment).
Role 6: Machine Learning Engineer (MLE)
Typical hiring companies: ML Platform teams at tech companies.
| Layer | Content |
|---|---|
| Key responsibilities | Build training/inference infrastructure; data and feature pipelines; model deployment, monitoring, rollback; optimize training and inference costs |
| Hard requirements | Solid software engineering (Python/C++); distributed systems experience; training framework familiarity (PyTorch); CI/CD and containerization |
| Bonus | Cloud platform familiarity (AWS/GCP); GPU cluster scheduling; MLOps toolchains (Kubeflow/W&B) |
Linked pages: MLOps & Model Deployment + How to Choose Frameworks & Tools.
Role 7: Research Scientist (Deep Learning)
Typical hiring companies: FAANG and AI lab research departments.
| Layer | Content |
|---|---|
| Key responsibilities | Explore frontier problems and publish papers; build large-scale experiments and benchmarks; collaborate with engineering teams for deployment |
| Hard requirements | PhD preferred; track record of top-tier publications; exceptional math and experimental design skills; ability to independently select research topics |
| Bonus | Open-source influence; cross-domain expertise (RL / multimodal / scientific computing); codebase maintenance experience |
Linked pages: Getting Started (papers) + Classic Paper Deep Dives.
Role 8: Applied Scientist
Typical hiring companies: AI product teams at e-commerce and cloud service providers.
| Layer | Content |
|---|---|
| Key responsibilities | Deploy latest methods into product capabilities; LLM applications (generation, summarization, retrieval); design evaluation systems and data strategies |
| Hard requirements | Master's / PhD; solid ML foundation and coding skills; end-to-end experience from research to deployment; strong communication and product sense |
| Bonus | LLM fine-tuning and evaluation experience; multimodal / speech and other vertical domain experience; dual background in both papers and engineering |
Linked pages: Large Language Models (LLMs) + Evaluation in Practice.
4. JD Skill Keyword Frequency Table
Broad word-frequency statistics across the 8 JDs above (data current as of 2026-08, verify before citing). High-frequency words represent "what the market keeps asking for":
| Skill Keyword | Frequency (out of 8 JDs) | Category |
|---|---|---|
| PyTorch | 8 | Framework |
| Transformer / Attention | 7 | Architecture |
| Distributed Training | 6 | Engineering |
| Fine-tuning / Alignment | 6 | LLM |
| Multimodal | 5 | Direction |
| Data Augmentation / Data Processing | 5 | Data |
| Evaluation Metrics / Experiments | 5 | Evaluation |
| Inference Optimization / Deployment | 4 | Engineering |
| RAG | 4 | LLM |
| Reinforcement Learning | 3 | Direction |
| Graph Neural Networks | 2 | Direction |
| CUDA | 2 | Low-level |
Takeaway: Regardless of specialization, "PyTorch + Transformer + Distributed + Evaluation" form the common foundation. Focus your energy on these four first for the highest return. The mapping from skill keywords to knowledge points is on the JD Knowledge Map page.
5. Salary and Degree Threshold Notes
Rough reference, data current as of 2026-08, verify before citing:
| Role Type | Domestic Salary (total package, ¥100K) | Overseas Salary (total package, $100K) | Degree Threshold |
|---|---|---|---|
| Deep Learning Researcher | 50–150 | 20–40 | Mostly PhD; Master's + strong papers negotiable |
| LLM Engineer | 45–120 | 18–35 | Mostly Master's; PhD a plus |
| Algorithm Engineer (General) | 35–90 | 15–30 | Mostly Master's |
| CV / NLP Engineer | 30–85 | 15–28 | Mostly Master's; Bachelor's + rich experience negotiable |
| ML Engineer | 40–100 | 18–35 | Bachelor's or Master's, engineering-focused |
| AI Infra Engineer | 50–130 | 20–40 | Bachelor's or Master's, system-focused |
Three reminders
- Salary ranges are rough estimates — variance within the same role is huge. Stock, performance bonuses, and team differences often outweigh average differences;
- Degree is a "filter," not a "ceiling": Research roles typically require a PhD, but engineering roles (ML / AI Infra) value code and system experience more — a Bachelor's can still command high salaries;
- "3 years of experience" on a JD can often be partially substituted with "high-quality projects + top-tier papers." Don't be deterred by literal thresholds. See Capability Match: What to Highlight in Your Resume.
6. How to Use This List for Job Hunting
- Pick 2–3 directions from the job landscape page;
- Search for those directions on target company websites, download 5–10 real JDs, and use the three-layer method to extract skill keywords;
- Cross-reference with the 8 breakdowns above — prioritize common skills (PyTorch / Transformer / Distributed / Evaluation);
- Feed the breakdown results into the JD Knowledge Map page to generate a study order.
Further Reading
- Module Overview & Job Landscape — Role categories and three-dimensional decision checklist
- JD Knowledge Map — Skill keywords → knowledge points → study plan
- Capability Match: What to Highlight in Your Resume — Write JD requirements into your resume
- Interview Question Bank — Prepare interviews based on high-frequency JD keywords
- Learning Paths: Three Routes — How to systematically fill knowledge gaps
- MLOps & Model Deployment — Core knowledge area for engineering roles