A hands-on, project-based guide to Machine Learning Operations built specifically for DevOps, Platform, and SRE engineers.
No ML background required. Every concept is explained through DevOps analogies you already understand.
If you are completely new to MLOps, read our DevOps to MLOps guide first.
Hit the Star! ⭐ If you are planning to use this repo for learning MLOps, please hit the star. Thanks!
- Who This Is For
- What We Build
- Prerequisites
- Phase 1: Local Dev & Pipelines
- Phase 2: Enterprise Orchestration for ML
- Learning Path
- Tech Stack
- Recommended Reading
- License
Most MLOps resources are written for data scientists learning infrastructure. This repo flips that.
You do not need to become a data scientist. But just like understanding how a Java application is built makes you a better DevOps engineer, understanding how an ML model is built, trained, and served makes you effective at operating ML workloads in production.
We'll start with the **basics of building and training a model, then work our way up to production-ready MLOps.
Just like a DevOps engineer doesn't write the application but understands how it is built and deployed, an MLOps engineer doesn't need to be a data scientist. Understanding the ML workflow helps you build, automate, deploy, and troubleshoot ML systems effectively.
Everything in this roadmap runs on Kubernetes and Docker, and tools you'll use in real-world MLOps platforms.
Goal: Build the required ML foundation by building an Employee attrition prediction model from your local systems.
Use case throughout: Employee attrition prediction for a large organisation (~500,000 employees). One problem, end-to-end. Keeps the focus on infrastructure and operations, not data science theory.
| Step | Title | Guide |
|---|---|---|
| 1 | Project Dataset Pipeline | Read the Guide |
| 2 | Data Preparation Stages | Read the Guide |
| 3 | Training & Building the Prediction Model | Read the Guide |
| 4 | From Model to Live API with KServe | Read the Guide |
Code: phase-1-local-dev/
Goal: Replace local, manual ML workflows with production-grade orchestration. Versioned data, automated pipelines, experiment tracking, and scalable training.
| Step | Title | Guide |
|---|---|---|
| 1 | Data Versioning Fundamentals | Read the Guide |
| 2 | Data Version Control (DVC) with AWS S3 | Read the Guide |
| 3 | Data Versioning using Airflow on Kubernetes | Read The Guide |
| 4 | Feature Store Fundamentals Explained | Read The Guide |
| 5 | Hands-on Feature Store with Feast on Kubernetes | Read The Guide |
| 6 | Kubeflow Explained for MLOps | Read The Guide |
| 7 | Hands-on Kubeflow on Kubernetes | Read The Guide |
| 8 | Kubeflow Trainer Explained (Hands-on) | Read the Guide |
| 9 | MLflow: A Practical Guide to Experiment Tracking | Read the Guide |
| 10 | KServe for MLOps: A Practical Guide | 🔜 Coming Next |
| 11 | Model Monitoring Explained | 🔜 Planned |
| 12 | Model Monitoring - Hands On | 🔜 Planned |
Code: phase-2-enterprise-setup/
In this capstone project, you'll build a production-style MLOps platform on Kubernetes by combining everything you've learned throughout this series.
By the end, you'll have built an enterprise-grade MLOps workflow that mirrors how modern organizations develop, train, track, and operate machine learning models on Kubernetes.
| Phase | Track | Title | Status |
|---|---|---|---|
| 1 | 🤖 Traditional ML | Local Dev & Pipelines | ✅ Done |
| 1 | 🤖 Traditional ML | K8s Deploy & Model Serving | ✅ Done |
| 3 | 🤖 Traditional ML | Enterprise Orchestration | 🔄 In Progress |
| 4 | 🤖 Traditional ML | Monitor & Observe | 🔜 Planned |
Here is the tech stack you will be using in this setup.
| Category | Tools |
|---|---|
| Data Pipeline | Python, Airflow |
| Model Training | scikit-learn |
| API / Serving | FastAPI, Flask, Docker, KServe |
| ML Orchestration | Kubeflow, MLflow Pipelines |
| Monitoring | Prometheus, Grafana, Evidently AI |
| Infrastructure | Kubernetes, Helm, GitHub Actions |
- Ray: Open-source distributed computing framework For Python & AI Workloads
- CML: CI/CD for Machine Learning Projects
- Dagster: Cloud-native data pipeline orchestrator
- Kestra: Open-source orchestration platform for data, AI, and infrastructure workflows
- Flyte: AI orchestration in pure Python
Dual licensed:
- Code (scripts, configs, manifests) — Apache 2.0
- Content (README, guides, docs) — All Rights Reserved
For commercial licensing: contact@devopscube.com