What should an ML engineer put on a resume?

Models and AI features you built and shipped, each with an evaluation metric, the stack, and how it was served or used. Add a production result such as a conversion lift or cost saving when you have one.

How do I write about LLM projects on a resume?

Describe the system, not the prompt: what it did, what it retrieved from, how you evaluated it, how you handled failures, and what it cost or saved. An evaluation number makes the bullet credible.

Should I include Kaggle competitions?

Yes if you're early in your career or the result was strong. List them under Projects with the ranking and the approach in one or two bullets.

What machine learning projects should I put on a resume?

Projects that go past a notebook: a model trained on real data, compared to a baseline, and served behind an API or app. Say what the metric was and what the result means.

What skills should a machine learning resume list?

Python, an ML framework (PyTorch or TensorFlow), scikit-learn, SQL, experiment tracking and deployment tools (MLflow, Docker, a cloud ML service), plus LLM tooling if the role needs it.

How to write a machine learning engineer resume

By Mubeen Jawed, founder of ResumeForDev · Updated November 7, 2026

Short answer

A machine learning engineer resume should show models and AI features that reached production and what they changed: an offline metric that improved (accuracy, recall, NDCG), an online result (conversion, cost, time saved), and how it was served (latency, scale). Name the stack in each bullet, such as PyTorch, scikit-learn, Hugging Face, an LLM API, vector search or MLOps tooling. Separate research or coursework from shipped systems so a reviewer can see which is which.

Key points

Pair an offline metric (F1, AUC, NDCG) with a production result where you have one.
Show the serving side: latency, throughput, cost per request, monitoring.
For LLM work, describe evaluation and guardrails, not only the prompt.
Label research, Kaggle and coursework as projects, not experience.

What ML hiring managers look for

ML engineering sits between research and backend engineering, and the resume has to show both halves. Reviewers want evidence you can improve a model (with a real evaluation, not a vibe) and evidence you can ship it: data pipelines, serving, monitoring and iteration. A notebook result alone reads as data science. A deployment without an evaluation reads as backend.

Before and after

Before:Built a machine learning model to predict customer churn.

After:Trained a gradient-boosted churn model (XGBoost, 40 features) that raised recall from 0.58 to 0.74 at the same precision; served it as a daily batch job feeding the retention team's outreach list.

Offline metric, the stack, and how the model was used in production.

Numbers for ML and AI work

ML metrics and how to phrase them
Type	Examples	Example phrasing
Offline quality	Accuracy, F1, AUC, NDCG, BLEU, eval pass rate	raised F1 from 0.71 to 0.83 on a held-out set
Online result	Conversion, CTR, deflection rate, revenue	lifted click-through 9% in an A/B test
Serving	p95 latency, requests/day, GPU hours	served 2M predictions/day at p95 under 80ms
Cost	Inference spend, training cost	cut LLM cost per request ~60% with caching and a smaller model
Data	Training set size, labels produced	built a 50k-example labelled dataset

Example ML and LLM bullets

Built a retrieval-augmented support assistant (embeddings, pgvector, an LLM API) with an evaluation set of 400 real tickets; reached 82% answer acceptance before launch.
Fine-tuned a BERT classifier in PyTorch for ticket routing, raising macro-F1 from 0.66 to 0.81 and removing a manual triage step.
Moved model training from ad-hoc notebooks to a versioned pipeline (MLflow, Airflow), making every production model reproducible from a commit.
Cut inference cost ~60% by routing simple requests to a smaller model and caching repeated prompts, with no drop on the eval set.
Added drift monitoring for the fraud model's top features, catching a data-source change two days before it would have hit precision.

The skills section for ML roles

Languages: Python, SQL
ML: PyTorch, scikit-learn, XGBoost, Hugging Face Transformers
LLM and retrieval: LLM APIs, embeddings, vector search (pgvector, FAISS), evaluation harnesses
Data and MLOps: pandas, NumPy, Spark, Airflow, MLflow, Docker

On ResumeForDev, set your target role to “Machine Learning Engineer”. Without a job post, the keyword check compares your resume against a data and ML vocabulary. Paste a job description and the keyword check uses that post's terms instead; on Pro, the writer also tailors the bullets to it. Before writing, it asks follow-up questions about the numbers above, and it removes any number or skill you never gave it. Your first resume is free.

Frequently asked questions

What should an ML engineer put on a resume?: Models and AI features you built and shipped, each with an evaluation metric, the stack, and how it was served or used. Add a production result such as a conversion lift or cost saving when you have one.
How do I write about LLM projects on a resume?: Describe the system, not the prompt: what it did, what it retrieved from, how you evaluated it, how you handled failures, and what it cost or saved. An evaluation number makes the bullet credible.
Should I include Kaggle competitions?: Yes if you're early in your career or the result was strong. List them under Projects with the ranking and the approach in one or two bullets.
What machine learning projects should I put on a resume?: Projects that go past a notebook: a model trained on real data, compared to a baseline, and served behind an API or app. Say what the metric was and what the result means.
What skills should a machine learning resume list?: Python, an ML framework (PyTorch or TensorFlow), scikit-learn, SQL, experiment tracking and deployment tools (MLflow, Docker, a cloud ML service), plus LLM tooling if the role needs it.