Machine Learning Engineer Resume Keywords (PyTorch, LLMs, Hugging Face)

ML engineer postings have a short, sharp keyword set. In 174 Machine Learning Engineer roles from our 15 August 2026 snapshot of 8,991 public postings, “machine learning” itself is almost universal (82%) — and then the stack splits into training (PyTorch) and applied LLM work.

For the neighboring role, see data engineer skills (SQL/dbt, not PyTorch).

What these JDs name

Skill	Share of ML engineer postings
Machine Learning	82%
LLMs	60%
Python	59%
PyTorch	37%
Deep Learning	27%
TensorFlow	26%
NLP	20%
SQL	20%
AWS	18%
Spark	18%
GCP	15%
CI/CD	14%

Hugging Face (12%) and LangChain (10%) show up often enough to matter on applied-LLM JDs, but they are not the default stack. Mirror the framework in the posting.

PyTorch still leads TensorFlow in this slice. That does not mean TensorFlow is dead — GitHub still shows TensorFlow as the larger public repo — it means these particular JDs are written in PyTorch more often. Use the name the posting uses.

Cloud is a three-way split (AWS 18%, GCP 15%, Azure 13%). Do not default to AWS on an ML resume if the company is a GCP shop.

Two JD shapes (write two summaries)

Trainer / applied scientist. PyTorch, Python, distributed training, evaluation metrics. Bullets look like: “Fine-tuned Llama-class models in PyTorch on 8×A100; cut eval latency 40%.”

Platform / LLM app. Hugging Face, LangChain, retrieval, eval harnesses, a cloud. Bullets look like: “Shipped a retrieval service (Hugging Face embeddings, LangChain, GCS) used by 3 product teams.”

If you have both, lead with whichever the posting repeats. A blended skills list with no matching bullets reads as a course transcript.

Data scientist vs ML engineer

In the same corpus, 195 Data Scientist postings still want Python (64%) and SQL (60%) more than PyTorch (9%). Machine learning shows up in 47% of DS JDs; LLMs in 16%. If you are applying to both, keep two skills blocks. SQL-heavy DS JDs will bounce an ML-only resume even when the modeling work is real.

More on the DS/analyst side: Looker, Tableau, and warehouse keywords on the role page, plus this overall 2026 skill ranking.

How to verify before you apply

Run the posting through the keyword checker. Confirm PyTorch vs TensorFlow, the cloud, and whether they said LLM, generative AI, or a specific model family — then use their words for work you actually did.

Method: 174 ML Engineer and 195 Data Scientist titles from 8,991 public postings on 15 August 2026 (Arbeitnow, Greenhouse, Ashby, Google Careers, Lever, RemoteOK, Remotive). Curated keyword extraction on description text. Aggregates only.