Machine Learning Engineer Job Description You Can Post

A machine learning engineer (ML engineer) takes models into production: train or fine-tune, package, deploy, monitor, and retrain. Amazon Ads in Seattle posts a base of $143,700–$194,400 a year as of 24 September 2026. Copy the template below, or use it to check a posting.

Pay figures are employer-posted ranges and labeled surveys, not compensation advice. Confirm the live posting before you offer or accept.

Key Takeaways

• The job is production ML software. A data scientist ships insight; an AI engineer ships an app on someone else's model; you ship the model.
• Posted cash is not one number. Amazon Ads Seattle lists $143,700–$194,400 base. Intuit Staff in Mountain View lists $202,500–$274,000 base. OpenAI Integrity in San Francisco lists $287,000–$555,000 plus equity.
• There is no BLS code for this title. Data Scientists (SOC 15-2051) show a May 2025 median of $120,230 and +35% outlook for 2025–35. That analog is not an ML engineer wage.
• Mid-level Amazon Ads asks 3+ years of software engineering. There is no required license. Google's Professional ML Engineer exam is $200 and optional.

What a Machine Learning Engineer Actually Is

You are a production software engineer who owns the model in a live system. You train or fine-tune, wrap it, put it behind an API or batch job, watch it, and retrain when the data drifts.

Amazon Ads writes the partnership with scientists, data engineers, product managers, and software engineers into the req. You report into an ML, ads, or platform org, not a research lab, unless the posting says otherwise.

This is an employer title. The U.S. has no SOC named "machine learning engineer." Closest official buckets are Data Scientists (SOC 15-2051) and research scientists (SOC 15-1221).

Intuit drew the fence on 5 March 2026: data scientists do analysis, feature engineering, and model development; ML engineers deploy, scale, and maintain those models (TensorFlow or PyTorch, Docker, Kubernetes, Airflow). Chip Huyen puts the same split in Machine Learning Interviews: data science generates business insights; ML engineering turns data into products.

Kore1 splits AI engineer as someone who builds on foundation models someone else trained: APIs, retrieval, agents. You still build, train, and operate the model.

A data engineer owns the warehouse and pipelines. A research scientist owns unpublished problems and papers.

The only official occupational standard that names this title is UK ST1398 from Skills England (apprenticeship in revision): take ML models to production and monitor them. That is a UK analog, not a U.S. license.

What Does a Machine Learning Engineer Do All Day?

The Amazon Ads req wants production ML and GenAI for advertiser communications: structured and unstructured data with scientists, GPU and distributed training, reusable assets, monitoring, debugging.

NVIDIA Agentic Systems asks for workflows across data generation, eval, debugging, and training orchestration. The must-have is building ML systems and software platforms, not only models.

Staff work at Intuit is feature pipelines, train and deploy, MLOps, A/B tests, and watching production models next to AI scientists.

GitLab writes "develop and implement models to support and enable experimental, beta, and generally available AI features," then integrate with backend, frontend, and fullstack. Intermediate and above join on-call.

Duties that show up across those postings:

• Turn a product or ops problem into a training and serving plan.
• Build or fine-tune models in Python with PyTorch or TensorFlow.
• Construct the data path: features, labels, joins, quality checks.
• Package and deploy (containers, orchestration, batch or online inference).
• Monitor live models: latency, drift, failures, cost.
• Retrain, roll back, and debug with scientists and software engineers.
• Run tests and experiments on the serving path.

Small-company postings pile software, data, and DevOps onto one person. Large-company seats shrink to small optimizations on an existing ranking product. Write the scope you actually need.

If the person only writes papers or a PoC, you wrote a research-scientist req. If they only move data and never own the model, you wrote a data engineer req.

Machine Learning Engineer Salary in 2026

Start with the employer's posted cash, and name the metric, city, and date. Do not blend these rows into one wage.

Amazon Ads also lists sign-on and RSUs in the package language. OpenAI lists equity on top of cash. Those are not the same metric as base.

Official U.S. analogs, labeled as other occupations:

• Data Scientists (SOC 15-2051): May 2025 median $120,230 ($57.80/hour); 275,600 jobs in 2025; outlook +35% for 2025–35 (+95,400 jobs, about 24,800 openings a year). BLS lists bachelor's as the entry education.
• Research scientists (SOC 15-1221): May 2025 median $140,300; 38,600 jobs; +22%. BLS lists master's as the entry education.
• Software developers: May 2025 median $135,980.

Named aggregators, labeled by method, never as the lede:

• Levels.fyi title median total compensation $280,000 (25th $200,000 / 75th $385,000 / 90th $498,000), last updated 24 September 2026. Self-report TC, not base.
• Indeed U.S. average base $186,730 from 5.1k salaries (low $110,574 / high $315,337), updated 14 September 2026. Job-board scrape.
• ZipRecruiter average $128,769 as of 22 June 2026 (median $125,200). Employer-posting mix, far below Indeed and Levels.

Pick the row that matches the offer you are writing.

Demand, not pay: the World Economic Forum WEF 2025 ranks AI and machine learning specialists among the three fastest-growing jobs in percentage terms through 2030, from a survey of 1,000+ large employers. That is not a U.S. SOC and not a salary.

Must-Haves Versus Nice-to-Haves

Mid-level Amazon Ads must-haves: 3+ years of software engineering outside internships; 2+ years of design and architecture; one programming language; ML, data mining, information retrieval, stats, or NLP. Nice-to-have on that req: 3+ years of full SDLC, a bachelor's in CS or equivalent, 1+ years of large-scale ML infra for recs, ads, or search.

Intuit Staff raises the bar to 7+ years, Python plus a data stack, SQL, CS fundamentals, Git, scale to millions of users, and AWS or GCP.

The NVIDIA Agentic Systems req wants expert Python, deep PyTorch, and 12+ years of software engineering. That is a systems seat, not the mid-level template.

GitLab publishes the public ladder (Associate 1+ years of ML or a Master's/PhD focused on ML; Intermediate 2+; Senior 3+; Staff 5+ or a PhD), requires Python, and puts Intermediate and above on-call. GitLab is a fully remote organization.

Must-have (the person cannot do the job without these):

• Production Python.
• PyTorch or TensorFlow.
• Ability to deploy and monitor a model.
• Software engineering years that match the seat: 3+ for mid-level Amazon Ads, 5–8 for senior/staff, 12+ on NVIDIA Agentic.

Nice-to-have:

• Google Professional ML Engineer ($200 exam). Vendor product, not a license.
• PhD, clearance, C++, ads or driving or safety domain, Java or R, a graduate degree, AWS or GCP.

A bachelor's in CS or a related field is the common entry line on U.S. postings. No U.S. license exists for this title. Skip generic skill dumps (reinforcement learning, audio/video processing) unless that work is on this req.

A Job Description You Can Paste

Company intro: [Company] builds [product / platform] that customers use in production. We hire machine learning engineers to take models from prototype to a monitored service.

Overview: You will own production ML for [team / product]. You train or fine-tune, package, deploy, monitor, and retrain. You work with scientists, data engineers, and software engineers on [City / remote].

Responsibilities:

• Translate a product problem into a training, evaluation, and serving plan.
• Build or fine-tune models in Python with PyTorch or TensorFlow.
• Build the data path: features, labels, quality checks, reusable datasets.
• Package, deploy, and scale training and inference ([GPU / distributed training] where this team uses it).
• Monitor live models for quality, drift, latency, and cost; debug failures.
• Partner with applied scientists, data engineers, product managers, and software engineers.
• Run tests and experiments on the production path.

Qualifications:

Must-have:

• [3+] years of software engineering (non-internship) for this mid-level seat.
• [2+] years designing and architecting production systems.
• Production Python and one of PyTorch or TensorFlow.
• Experience deploying and monitoring models.
• One of ML, data mining, information retrieval, statistics, or NLP.

Nice-to-have:

• [1+] years of large-scale ML infrastructure (recommendations, ads, search, or equivalent).
• Cloud (AWS or GCP), containers, and workflow orchestration.
• Google Professional ML Engineer certification.
• Domain depth in [ads / search / safety / robotics].

Hours and benefits: [schedule]; location [City / remote]; posted pay [salary range]; package [benefits].

How to apply: [how to apply]

NVIDIA Agentic lists BS/MS or equivalent next to a 12-year software bar. Do not paste that bar onto a 3-year ads req.

Amazon Ads Machine Learning Engineer (Filled Example)

Company intro: Amazon Ads Advertising & Marketing Performance Intelligence (AMPI) publishes automated marketing communications for advertisers. The team uses machine learning, large language models, and specialized agents so those communications stay cohesive, relevant, and personalized at scale.

Overview: You own production ML and GenAI for advertiser communications on Amazon Ads in Seattle. You process structured and unstructured data with scientists, scale training infrastructure, and keep models monitored in production.

Responsibilities:

• Build production ML and GenAI for advertiser communications.
• Process structured and unstructured data with applied scientists.
• Scale ML and LLM infrastructure (GPU, memory, distributed training).
• Ship reusable training and serving assets other teams can pick up.
• Monitor and debug live models.
• Partner with applied scientists, data engineers, product managers, and software engineers.

Qualifications:

Must-have:

• 3+ years of software engineering outside internships.
• 2+ years of design and architecture.
• One programming language used in production.
• ML, data mining, information retrieval, statistics, or NLP.

Nice-to-have:

• 3+ years of full software development lifecycle.
• Bachelor's in computer science or equivalent.
• 1+ years of large-scale ML infrastructure for recommendations, ads, or search.

Hours and benefits: Seattle, WA. Posted base $143,700–$194,400 a year as of 24 September 2026. Package language on the live req includes sign-on and RSUs.

How to apply: Apply now on Amazon.jobs.

Titles Hiring Managers Mix Up

• Data scientist: insights and model prototypes. You productionize. Intuit and Chip Huyen both use that split.
• AI engineer: applications on top of models and agents, not training the core model (Kore1).
• Data engineer: warehouse, ETL, and the data platform. You own the model once it serves.
• Software engineer: applications. You ship models into those systems.
• MLOps engineer: CI/CD, infra, and monitoring of models already built. ST1398 lists overlapping titles; keep ownership of the model on this req if that is the job.
• Research scientist: unpublished problems, papers, PoC. Production sits outside the remit.
• Applied scientist: Amazon and lab dialect for a modeler who ships, with more modeling than this IC.
• Engineering manager, machine learning: people-manager of MLE teams. GitLab splits that track. Not the IC role.

A customer-embedded software seat is a forward deployed engineer, not this occupation.