Getting a model to work once is data science. Keeping it trained, versioned, served, watched and retrained is MLOps. Learn the whole line: experiment tracking, feature stores, pipelines, serving, drift monitoring, CI/CD for ML and Kubernetes, mostly on AWS.
Counted from the lesson files. 98 chapter assessments have 30 questions, one has 25 and one has 12, mixing multiple choice, short answers, scenarios and code debugging.
A data scientist hands over a better model. Before it carries real traffic, an MLOps engineer checks it against what the business actually requires, not just its accuracy. Here are the requirements from the course's three capstones. Pick one, see what's holding the candidate, and fix it.
Requirements are from the capstone briefs in Lesson 12. The candidate models and their numbers are illustrative.
The same skills are hired under several titles. Each board shows the lines (lessons) that stop there.
A Fundamental-tier bootcamp: expect the first level of these titles, coming from DevOps or from data science.
MLOps is the interchange between DevOps and machine learning. You need one of the two already; the bootcamp teaches the other half. Tick what you've actually done.
Each lesson is a line and each chapter a station. Where two chapters cover the same ground from different sides, you can change lines. Pick a line on the map, then a station.
12 chapters setting up the toolchain, ending with Killercoda, a free browser sandbox for Kubernetes and container practice.
Three companies, three phases each, and a reference solution. See them ↓
Eight modules that turn the capstones into case studies, a model card and an experiment report.
Data Drift Detection: five written sections, a practice set of five assignments and a 30-question assessment. Then try the chapter's most useful warning for yourself.
A credit model's input feature, split into 10 equal-frequency bins on the training data, so each bin should hold 10%. Mobile users are 10% of traffic. Drift their distribution and compare the two monitors.
PSI = Σ (p − q) ⋅ ln(p / q), with a small epsilon for empty bins. Bands from Section B: under 0.10 stable, 0.10–0.20 investigate, over 0.20 significant, over 0.25 major.
Each capstone is a company with real constraints, worked through in three phases: data and feature assessment, pipeline and deployment design, then monitoring and rollout. The reference solution compares reasoning, not answers: four decisions, each with the option chosen and the alternative set aside.
Eight modules make your capstone work something a hiring manager can judge, with guidance on anonymising anything confidential first.
Gross annual base salary. Each country is a line running from a first MLOps role to senior; stations mark the sourced figures it's based on. Germany has real MLOps data; for the Netherlands and Belgium we read across related roles.
Self-paced doesn't mean on your own. Mentor Bob is an AI study assistant in the corner of every section. It has already read the section you're on, so you can ask about it in your own words.
Yes, if you want your models to reach production and stay there. Expect the infrastructure lessons to be the steepest: containers and infrastructure as code (Lesson 1), CI/CD for ML (Lesson 7) and Kubernetes (Lesson 8). Lesson 11 walks you through installing the toolchain.
Enough to read Python. The course doesn't teach model theory; it teaches what you need to run models: training infrastructure, tuning, experiment tracking and registries in Lesson 2, and what drift and model quality mean in Lesson 6. Expect those to be the steepest lessons for you.
MLOps runs machine-learning models in production: training, serving, monitoring and retraining them. AIOps uses data, ML and automation to run IT systems: anomaly detection, alert noise, automatic remediation. For the second, see AIOps Fundamental.
It's AWS-focused: SageMaker (training, endpoints, pipelines, Feature Store, Model Monitor), EKS, Step Functions. It also covers the open tools teams use with it or instead: MLflow, Kubeflow, Airflow, Feast, KServe, Seldon, Triton, Ray, DVC, Evidently, ONNX and Argo, with comparisons between them.
It's written material, but every chapter has a practice section built around realistic scenarios, with code and configuration to work through, plus an assessment with explained answers. The three capstones are where you design a whole system.
Where they matter to operations, such as inference optimisation, GPU cost and serving. But the course is about running ML models in general, not building LLM applications. For that, see LLM Engineer for Beginners.
That depends on your pace and background, so we don't promise a number of weeks. For scale, there are about 2.5 million words across the lessons, code included. Access is for life, so there's no deadline.
Every lesson unlocks straight away in your dashboard, with Mentor Bob in each section. It's a one-time payment with lifetime access, so no subscription and no renewal.
All 10 lessons and 100 chapters, plus the installation guide, three capstones and the portfolio lessons, unlocked as soon as you enroll.
✓ You already own this bootcamp. It's waiting in your Active Bootcamps.