MLOps Engineer jobs: skills, pay and how to get hired
Contents
See open MLOps Engineer jobsScore your CV for these jobs
What an MLOps engineer does
An MLOps engineer builds and runs the machinery around machine learning models: the pipelines that prepare data and train models, the registry that keeps track of every version, the services that answer requests, and the monitoring that says when a model has gone stale. The aim is that a data scientist's model can go from a notebook to production, and back for retraining, safely and without heroics. Ads use MLOps Engineer, ML Platform Engineer, ML Infrastructure Engineer and, in German ads, MLOps Ingenieur.
The neighbours split the work differently. Data scientists find the signal and build the models; the AI engineer builds products on machine learning and large language models; the data engineer owns the data pipelines; the DevOps engineer runs the delivery of software in general, and the platform engineer the internal platform every team builds on. MLOps sits where these meet, for models. The steps from junior MLOps engineer to head of ML platform are in the career paths of this role.
Rules that touch models differ by place, and each country's section names its own. This page describes the work of running models; it is not legal or career advice.
The skills the ads ask for
We read a sample of the ads every day and count the skills they name. On a computer, hover a skill in the text to see its share; on a phone, the table has the same figures.
| Skill | What is expected |
|---|---|
| MLOps | Run the whole life of a model in production: build, test, release, watch and retrain, with as little hand work as possible. |
| Kubernetes | Run training jobs and model services on Kubernetes, with sensible limits, scaling and health checks. |
| Python | Write the glue of the platform, pipeline steps and tests in Python, and read the data scientists' code with ease. |
| Docker | Package a model and its dependencies into an image that runs the same on a laptop and in production. |
| Machine Learning | Understand how models are trained and evaluated well enough to spot a broken metric or a leak. |
| CI/CD | Test and release models and pipelines through the same automated checks as other software. |
| ML Platforms | Use a managed ML platform, such as SageMaker, Vertex AI or Azure ML, or Kubeflow, and know what it does for you. |
| AWS | Build ML infrastructure on AWS: storage, compute, permissions and the services a model needs. |
| MLflow | Track experiments and register models in MLflow, with the parameters, data and metrics that explain each one. |
| Terraform | Describe the platform's infrastructure as code, so that environments can be rebuilt and reviewed. |
| Model Serving | Serve models in batch or in real time, within the latency and cost the product can carry. |
| Model Monitoring | Watch a live model's inputs, outputs and quality, and raise an alarm when they move. |
The core is MLOpsRun the whole life of a model in production: build, test, release, watch and retrain, with as little hand work as possible. practice: CI/CDTest and release models and pipelines through the same automated checks as other software. for models, containersPackage a model and its dependencies into an image that runs the same on a laptop and in production., KubernetesRun training jobs and model services on Kubernetes, with sensible limits, scaling and health checks., infrastructure as code with TerraformDescribe the platform's infrastructure as code, so that environments can be rebuilt and reviewed., and pipelines scheduled in tools such as AirflowSchedule and watch data and training pipelines, with retries and alerts that make sense.. Around it come model servingServe models in batch or in real time, within the latency and cost the product can carry., model monitoringWatch a live model's inputs, outputs and quality, and raise an alarm when they move. and, in larger teams, feature storesKnow when a feature store helps a team reuse features and keep training and serving in step.. Ads want PythonWrite the glue of the platform, pipeline steps and tests in Python, and read the data scientists' code with ease. and enough machine learningUnderstand how models are trained and evaluated well enough to spot a broken metric or a leak. and PyTorchRead and run PyTorch training code, and package its models for serving. to understand what the data scientists hand over.
The platforms are named side by side and none is a must everywhere: MLflowTrack experiments and register models in MLflow, with the parameters, data and metrics that explain each one. for experiments and the model registry, and Kubeflow, SageMaker, Vertex AI and Azure ML as ML platformsUse a managed ML platform, such as SageMaker, Vertex AI or Azure ML, or Kubeflow, and know what it does for you., each tied to its cloud, AWSBuild ML infrastructure on AWS: storage, compute, permissions and the services a model needs., Google CloudBuild ML infrastructure on Google Cloud, with its storage, compute and managed services. or AzureBuild ML infrastructure on Azure, with its identity, storage and compute services., except Kubeflow, which runs on any Kubernetes. Much of the job is explaining trade-offsExplain a platform choice or an incident to data scientists, engineers and managers in plain words. to data scientists and product teams, in EnglishThe language of the tools, the documentation and most international ML teams. and, in many German teams, GermanAsked for in many German ads, for teams and stakeholders in Germany..
Where the jobs are, country by country
Each country below has its own figures: which cities have the most jobs in our list, whether the number is going up or down, how often ads offer remote or hybrid work, the titles used in the country, and pay ranges from published research, in the local currency.
Germany
Pay (Base salary)
- Entry level: €49,600 to €57,800 a year
- Mid-level: €57,800 to €74,786 a year
- Senior: €59,131 to €85,000 a year
Total pay: Payscale puts the yearly bonus of machine learning engineers in Germany at about €1,000 to €11,000.
No German source covers MLOps engineers alone, so these are machine learning engineers' figures. Entry level is the lower half of StepStone's range for machine learning engineers and mid-level runs from its average to its average for engineers with experience, both a salary site's estimate. The senior range runs from Payscale's German average for machine learning engineers to the salary below which it puts nine in ten, and rests on 157 salaries.
Sources: StepStone: Machine Learning Engineer (2026), Payscale: machine learning engineer (Feb 2026).
Pay researched on 29 Oct 2026.
Rules and work: under the EU AI Act, systems that count as high-risk, such as some used in hiring, credit or medical devices, need logging, documentation and human oversight, which lands on the MLOps team; the EU's GDPR covers the personal data in training sets. Some German teams also check cloud AI services against the AIC4 criteria of the BSI, the German federal office for information security. German ads mostly say MLOps Engineer, sometimes MLOps Ingenieur.
Language: many German ML teams work in English, while others ask for GermanAsked for in many German ads, for teams and stakeholders in Germany. next to EnglishThe language of the tools, the documentation and most international ML teams..
United States
Pay (Base salary)
- Entry level: $103,880 to $116,613 a year
- Mid-level: $116,613 to $139,115 a year
- Senior: $139,115 to $171,000 a year
Total pay: no source gives a bonus for MLOps engineers alone; Payscale puts the yearly bonus of US machine learning engineers at $3,000 to $23,000, with profit sharing of up to about $28,000 in some firms.
Entry level runs from the 10th to the 25th percentile of Salary.com's US range for MLOps engineers, and mid-level from there to its 75th percentile; both are a salary site's estimate. Senior runs from that 75th percentile to the salary below which Payscale puts nine in ten US machine learning engineers, the nearest role it covers.
Sources: Salary.com: MLOps engineer (Oct 2026), Payscale: machine learning engineer (Oct 2026).
Pay researched on 29 Oct 2026.
Rules and work: the US has no single AI law; many teams map their controls to the AI Risk Management Framework of NIST, a US federal standards agency, which is voluntary. US ads mostly offer full-time employment, in software, finance, health, retail and cloud companies.
United Kingdom
Pay (Base salary)
- Entry level: £52,700 to £63,750 a year
- Mid-level: £63,750 to £100,000 a year
- Senior: £100,000 to £115,000 a year
Total pay: no source gives a bonus for UK MLOps engineers alone; Payscale puts the yearly bonus of UK machine learning engineers at £2,000 to £22,000.
All three levels come from ITJobsWatch's salaries of UK jobs that cite MLOps, over the six months to 29 Oct 2026: entry level from its 10th to its 25th percentile, mid-level from the 25th to the 75th, senior from the 75th to the 90th. The jobs that cite MLOps include some machine learning and data engineering jobs.
Sources: ITJobsWatch: MLOps (Oct 2026), Payscale: machine learning engineer (Jul 2026).
Pay researched on 29 Oct 2026.
Rules: personal data in training sets falls under the UK GDPR, the UK's version of the EU's GDPR, and UK regulators apply their own rules to AI in their sectors rather than one AI law. UK MLOps work sits in finance, retail, health and software.
Switzerland
Pay (Base salary)
- Entry level: CHF 80,000 to CHF 101,152 a year
- Mid-level: CHF 87,024 to CHF 120,512 a year
- Senior: CHF 120,512 to CHF 166,887 a year
Total pay: Payscale puts the yearly bonus of Swiss machine learning engineers at up to about CHF 7,000.
No Swiss source covers MLOps engineers alone, so these are machine learning engineers' figures, and both pages rest on few salaries. Entry level runs from Payscale's Swiss average for machine learning engineers in their first year to its average for one to four years, including bonuses. Mid-level runs from the low end of Indeed's Swiss range for machine learning developers to its average, and senior from that average to its top.
Sources: Payscale: machine learning engineer (Sep 2026), Indeed: machine learning engineer (Oct 2026).
Pay researched on 29 Oct 2026.
Rules: personal data in training sets falls under the Swiss FADP, the Federal Act on Data Protection. Swiss MLOps work is often in banking, insurance and pharma, and Swiss ads often ask for German or French next to EnglishThe language of the tools, the documentation and most international ML teams..
India
Pay (Base salary)
- Entry level: ₹3.62 lakh to ₹5.96 lakh a year
- Mid-level: ₹6.19 lakh to ₹11.22 lakh a year
- Senior: ₹11.22 lakh to ₹20.32 lakh a year
Total pay: Payscale puts the yearly bonus of machine learning engineers in India at ₹26,000 to ₹5.56 lakh.
No Indian source covers MLOps engineers alone, so these are machine learning engineers' figures. Entry level runs from the lower tenth of Payscale's range to its average for engineers in their first year, including bonuses. Mid-level runs from the low end of Indeed's India range for machine learning engineers to its average, and senior from that average to its top; the Indeed page rests on few salaries.
Sources: Payscale: machine learning engineer (Jul 2026), Indeed: machine learning engineer (Oct 2026).
Pay researched on 29 Oct 2026.
Rules and work: India's Digital Personal Data Protection Act of 2023 covers the personal data that models are trained on. Many Indian MLOps roles are in the technology centres of global companies and in Indian product companies, often serving teams in other time zones.
The rules that touch models in production are named side by side, each with its place. In the EU, the EU AI Act asks for risk management, logging and human oversight of high-risk systems, and the EU's GDPR covers personal data in training sets; some German teams check cloud AI services against the AIC4 criteria of the German BSI; in the US, the NIST AI Risk Management Framework is voluntary guidance; the UK applies the UK GDPR and its sector regulators; Swiss teams follow the FADP; and India has its Digital Personal Data Protection Act. Which of them applies depends on the system and the data, and the company's lawyers decide it.
One model, from notebook to production
- Package: turn the data scientist's training code into a pipeline step with pinned versions and tests.
- Track: log every training run, its data, parameters and metrics, and register the model that passes the checks.
- Release: deploy the new version behind a canary or in shadow, compare it with the old one, then switch.
- Watch: follow latency, errors, input data and the quality of the predictions against what the model saw in training.
- Retrain: start a new run when the data has moved or quality drops, and go round again.
Most model failures in production are quiet: nothing crashes, the answers simply get worse. The job is to make that visible before the business notices.
Your CV for an MLOps engineer job
An MLOps engineer's CV first meets software that searches it for the ad's words, then a lead engineer who wants to know which models you put into production, how they were served and what you did when they drifted. Give both: the keywords and the evidence.
Use the keywords the ads use
- Write the work as the ads do: "MLOps", "model deployment", "Kubernetes", "CI/CD", "monitoring", and the platform you used. The skills table above shows which keywords the ads use most.
- Name the scale: how many models, requests a second or teams the platform served, as far as you may say.
- Say what was live: a model in production for real users counts for more than a demo.
- A cloud course or exam you are still taking goes on the CV as in progress, with the planned date.
Prove each skill
An MLOps line proves itself with a model, the system around it and what changed for the team.
| Skill | Weak line | Strong line, with proof |
|---|---|---|
| PipelinesRun the whole life of a model in production: build, test, release, watch and retrain, with as little hand work as possible. | Built ML pipelines. | Turned a hand-run training notebook into a scheduled pipeline that retrains and tests the model each week. |
| ServingServe models in batch or in real time, within the latency and cost the product can carry. | Deployed models. | Moved a pricing model to a real-time service that answers within the product's latency limit. |
| MonitoringWatch a live model's inputs, outputs and quality, and raise an alarm when they move. | Set up monitoring. | Added drift and quality checks that caught a broken upstream feed before customers saw wrong scores. |
| RegistryTrack experiments and register models in MLflow, with the parameters, data and metrics that explain each one. | Used MLflow. | Introduced experiment tracking and a model registry, so every model in production can be traced to its data and code. |
| PlatformRun training jobs and model services on Kubernetes, with sensible limits, scaling and health checks. | Worked with Kubernetes. | Ran training jobs on shared GPU nodes with quotas, which cut waiting time for the data scientists. |
Leave out what the employer keeps confidential: no customer data, no model details the company treats as secret.
How we read our data
- The jobs: the MLOps engineer jobs in our job search (MLOps engineers, machine learning operations engineers, ML platform and ML infrastructure engineers, and the German title MLOps Ingenieur), matched by their titles the way the search matches them; machine learning engineers, DevOps engineers and platform engineers are counted on their own pages. Our list is not the whole market, so we give shares and rankings, never counts.
- The skills: every day we read a sample of these ads and count the skills in them with a fixed word list. Only the counts are kept, never the text of an ad.
- Pay: from published salary research, each source named with its date, as a range from entry level to senior. Our ads are not used for it.
- Dates: every figure is worked out again each day and shows the day it was worked out.
Frequently asked questions
- What does an MLOps engineer do?
- Builds and runs the pipelines, services and monitoring that keep machine learning models working in production.
- Is MLOps the same as DevOps?
- Related: MLOps adds data, training, model versions and drift to the delivery work DevOps does for software.
- Do I need to be a data scientist?
- No, but you need enough machine learning to understand what a model needs and when its numbers look wrong.
- Which ML platform should I learn?
- The one your target ads name; MLflow, Kubeflow, SageMaker, Vertex AI and Azure ML share the same ideas.
- What should an MLOps engineer's CV show first?
- Models you put into production, how they were served and watched, and what changed for the team.
More on the MLOps engineer role
The cvscored team
See open MLOps Engineer jobsScore your CV for these jobs
Related roles
Jobs from Arbeitnow and the Bundesagentur für Arbeit