DataTalks.Club's Free ML Zoomcamp Hits 14,600 Stars and Opens 2026 Registration
DataTalks.Club's free four-month course on production ML engineering has crossed 14.6k stars, covering everything from regression to Kubernetes serving.
Takeaways
- DataTalks.Club's free ML Zoomcamp hits 14.6k stars, 2026 cohort starts September 14
- Four-month course covers regression, classification, trees, deep learning, and deployment
- Stack: scikit-learn, PyTorch, TensorFlow, FastAPI, Docker, Kubernetes, AWS Lambda
- Certificate requires two end-to-end projects plus peer reviews during the live cohort
- Prerequisites are one year of programming and command-line comfort, no prior ML needed
- Materials are free and self-paced anytime via the YouTube playlist
DataTalks.Club’s free, community-run Machine Learning Zoomcamp has returned to GitHub Trending after its repository passed 14,600 stars. Registration is open for the 2026 cohort, which starts September 14 and covers the path from a raw dataset to a containerized prediction service running on Kubernetes.
The curriculum follows trained models into deployment. Learners evaluate models, expose predictions through an HTTP API, package the application with Docker, and deploy it using serverless or container orchestration tools. That production focus addresses the work required to move a model out of a notebook and connect it to other software.
GitHub stars track interest in the repository. The public lessons, assignments, project criteria, and deployment examples make the course’s scope available for inspection before registration.
From notebook to running service
Python anchors the course, with NumPy and pandas for data preparation, scikit-learn for classical machine learning, and TensorFlow, PyTorch, and Keras for neural networks. The deployment modules introduce FastAPI, Docker, AWS Lambda, Kubernetes, TensorFlow Serving, and KServe.
| Stage | Topics and tools | Practical output |
|---|---|---|
| Problem framing | CRISP-DM, project setup, and reproducible workflows | A defined prediction task and structured project |
| Tabular models | Linear regression, logistic regression, regularization, categorical encoding, and feature selection | Regression and classification baselines |
| Evaluation | Precision, recall, F1, ROC/AUC, confusion matrices, and cross-validation | Model comparisons grounded in held-out performance |
| Tree ensembles | Decision trees, random forests, and gradient boosting with XGBoost | Models for structured data and nonlinear relationships |
| Deep learning | Neural networks, convolutional networks, and transfer learning with PyTorch, TensorFlow, and Keras | Image-classification systems |
| API deployment | Model serialization, FastAPI, Docker, and cloud deployment | A packaged prediction service callable by other applications |
| Serverless and orchestration | AWS Lambda, API Gateway, Kubernetes, TensorFlow Serving, and KServe | Serverless endpoints and container-based model serving |
FastAPI turns Python prediction code into an HTTP endpoint, while Docker bundles the code and its dependencies into a portable container. AWS Lambda runs prediction handlers without a dedicated server. Kubernetes manages deployed containers, and TensorFlow Serving and KServe provide specialized layers for hosting models at scale.
A cohort built around deadlines
Cohort participants watch prerecorded lectures on their own schedules and follow shared deadlines for homework and projects. The cohort also adds a leaderboard, peer review, and certificate eligibility.
| Registration | Open for the 2026 cohort |
|---|---|
| Start date | September 14, 2026 |
| Schedule | Modules run from September through December; final projects conclude in January |
| Cost | Free |
| Certificate | Available during the live cohort after two qualifying projects and the required peer reviews |
| Prerequisites | About one year of programming experience and comfort with a command line |
| Prior ML experience | None required |
| Hardware | No local GPU required; deep-learning exercises can run in cloud environments |
The public course repository and video playlist also support self-paced study. Self-paced learners retain access to the material, while certificates remain tied to cohort deadlines, project submissions, and peer review.
The certificate requires deployed work
Certificate candidates choose datasets, define prediction tasks, train and evaluate models, and deploy working web services. Each candidate must submit two qualifying projects and complete the assigned reviews of other learners’ work.
Published projects show the expected scope. One learner built a blood cell classifier to detect signs of acute lymphoblastic leukemia. Another trained an Xception-based classifier on approximately 15,000 waste images, reported 93.3% test accuracy, and served predictions through a Docker-packaged API.
Built for production-minded beginners
Developers with general programming experience can use the course to add model evaluation, API development, containerization, and cloud deployment to their portfolios. Public repositories and running services give reviewers concrete artifacts to inspect beyond a completion certificate.
The syllabus centers on supervised learning, image classification, model serving, and deployment infrastructure. Research topics such as large-language-model pretraining, reinforcement learning from human feedback, and diffusion-model internals fall outside its stated scope.
Kubernetes and serverless modules provide an introduction to production deployment rather than comprehensive operating experience. Long-lived systems also require security, monitoring, cost control, incident response, and maintenance practices that extend beyond a course project.