MLOps Zoomcamp maps the path to production ML

This free course shows how to build, deploy, and monitor ML systems end to end.

MLOps Zoomcamp is a free 9-week course with 14.8k GitHub stars that turns machine learning basics into production practice. If you want a guided path from experiments to monitoring, these seven parts show what each module covers and who it helps most.

Item	Focus	Notable detail
Experiment tracking	Model management	MLflow basics, registry, saving and loading
Orchestration	ML pipelines	Workflow automation
Deployment	Serving models	Online, streaming, and batch options
Monitoring	Service health	Prometheus, Evidently, Grafana, Prefect, MongoDB
Best practices	Engineering hygiene	Testing, linting, pre-commit, CI/CD, Terraform

1. Introduction and setup

The opening module explains what MLOps is, why it matters, and how the course is organized. It also uses the NY Taxi dataset as a running example, which makes the lessons concrete instead of abstract.

This is the right starting point if you know basic machine learning but have not yet connected it to production workflows. The course asks for Python, Docker, command line basics, and about a year of programming experience, so it assumes some comfort with tools, not just notebooks.

• Core topics: MLOps maturity model, course structure, environment setup
• Audience: data scientists, ML engineers, software engineers
• Format: pre-recorded lectures with homework

2. Experiment tracking and model management

This module covers the first pain point many teams hit: remembering which run produced which model. It introduces experiment tracking, MLflow, model saving and loading, and the model registry, so you can compare runs and keep models organized.

If you have ever lost track of hyperparameters, metrics, or model versions, this section gives a practical fix. It is especially useful for anyone moving from one-off training scripts to a repeatable workflow.

Topics: experiment tracking, MLflow basics, model registry, hands-on exercises, homework

3. Orchestration and ML pipelines

Once experiments are under control, the course moves to orchestration. This module focuses on workflow orchestration and ML pipelines, which is the step that turns scattered scripts into a coordinated process.

That matters when training, validation, and data preparation need to run in a reliable order. For teams that want fewer manual steps and less fragile glue code, this is one of the most practical sections in the syllabus.

• Goal: automate multi-step ML workflows
• Outcome: reusable pipeline structure
• Homework: build an orchestration exercise

4. Deployment strategies

The deployment module compares online, streaming, and offline serving, then shows how to ship models with Flask, AWS Kinesis, Lambda, and batch scoring. That mix is useful because not every model needs the same serving pattern.

Readers who mainly know training will get a clearer picture of how models actually reach users and systems. The module helps you choose between a web service, a streaming setup, or batch processing based on the job at hand.

• Online deployment: web service or streaming
• Offline deployment: batch scoring
• Tools mentioned: Flask, AWS Kinesis, Lambda

5. Monitoring and service health

Monitoring is where production ML gets real, and this course treats it as a full module rather than an afterthought. It covers service monitoring with Prometheus, Evidently, and Grafana, plus batch monitoring with Prefect, MongoDB, and Evidently.

That split is important because web services and batch jobs fail in different ways. If you want to watch data drift, service behavior, and job outcomes, this module gives a broad starting toolkit.

Web monitoring: Prometheus + Evidently + Grafana Batch monitoring: Prefect + MongoDB + Evidently

6. Best practices for shipping ML

This module shifts from ML-specific tasks to the engineering habits that keep systems maintainable. It includes unit and integration testing, linting, formatting, pre-commit hooks, CI/CD with GitHub Actions, and infrastructure as code with Terraform.

For teams that already have a model but struggle with quality control, this is the section that ties everything together. It helps you build workflows that are easier to review, automate, and reproduce.

• Testing: unit and integration
• Automation: GitHub Actions, pre-commit
• Infrastructure: Terraform

7. Final project and certificate path

The final project asks you to combine the earlier modules into an end-to-end MLOps pipeline. That makes the course more than a video series, because you finish with a portfolio-ready system rather than isolated exercises.

Certificates are available for learners who complete the final project during a live cohort, but the course is now fully available for self-paced study. The repository also points to Slack, Telegram, and the FAQ for support, which makes it easier to keep moving when you get stuck.

How to decide

Pick this course if you want a free, structured introduction to production ML and you already know enough Python and Docker to follow along. It is strongest for learners who need a practical map from experimentation to deployment and monitoring.

If you mainly want a certificate, the live-cohort route matters. If you want the skills, the self-paced path is enough: the materials, homework, and final project are all there in the repository.