How to use the MLOps Roadmap
- 1. Tick what you knowOpen each stage and tick the topics you can already do confidently.
- 2. Pick your paceChoose how many hours a week you can study; the estimate updates for what's left.
- 3. Get your planSee time left per stage and your next three topics, and download the checklist.
Who this roadmap is for
MLOps applies software engineering and DevOps practices to machine learning. Models depend on data that changes, so shipping one model is the easy part. Keeping it accurate, reproducible and cheap to run is the job.
This roadmap assumes you know basic ML. If you don't, start with the machine learning roadmap and come back once you have trained a few models.
The 8 stages
Each of the 8 stages of the MLOps Engineer path ends with a milestone project, because building something is what turns topics into skills.
- Foundations (about 4 weeks): Python packaging, virtual environments and clean code, Git, branching and code review, Linux and shell scripting and more. Milestone: Turn a notebook into a tested Python package.
- Data management (about 3 weeks): Data versioning (DVC or lakeFS), Data validation (Great Expectations or pandera), Feature stores idea and more. Milestone: Version a dataset and add validation checks to a pipeline.
- Experiment tracking (about 2 weeks): Tracking runs, parameters and metrics with MLflow or W&B, Model registry and stages, Reproducible environments and more. Milestone: Track 10 training runs and register the best model.
- Pipelines and orchestration (about 3 weeks): Pipeline tools: Airflow, Prefect or Kubeflow Pipelines, Scheduling retraining, Caching and idempotent steps and more. Milestone: Automated pipeline: ingest → validate → train → evaluate.
- Containers and serving (about 4 weeks): Docker images for ML, Model serving: FastAPI, BentoML, or a model server, Batch vs real-time inference and more. Milestone: Serve your registered model in a container with health checks.
- CI/CD for ML (about 2 weeks): Tests for code, data and models, CI pipelines with GitHub Actions, Continuous training and deployment gates and more. Milestone: CI that blocks a model with worse metrics from shipping.
- Monitoring (about 2 weeks): Latency, errors and throughput, Data drift and concept drift, Alerting and retraining triggers and more. Milestone: Drift dashboard with an alert rule.
- LLMOps and job ready (about 3 weeks): Prompt and model versioning for LLM apps, Evaluating LLM outputs, Cost and latency tracking and more. Milestone: End-to-end MLOps project write-up with a diagram.
How the time estimate works
Worked example: with nothing ticked and 10 hours a week, the plan shows about 23 weeks for all 33 topics. Choose 20 hours a week and it becomes about 11.5 weeks. Tick every topic in the first stage, Foundations, and that row shows "done" while the total at 10 hours a week drops by 4 weeks to about 19.
Week counts are our suggestion for steady self-study of the MLOps Engineer path, based on how much each stage covers. They are not survey data or a promise; use them to plan, not to judge yourself.
Tips for following it
- Make every training run reproducible: same code version, data version and environment.
- Add a metric gate so a worse model can't reach production automatically.
- Monitor inputs as well as outputs; data drift often shows up before accuracy drops.
Frequently asked questions
Which tools should I learn?
Tools change; the roadmap names common options (MLflow, DVC, Airflow, Prefect) but the concepts matter more than any one tool.
Is MLOps a separate job?
Some companies hire MLOps or ML platform engineers; in others, ML engineers or DevOps engineers cover these tasks.
Where are my ticks saved?
Your MLOps Roadmap ticks are kept only in this browser on this device, so nothing you tick is sent anywhere. Clearing site data or switching browsers starts you fresh; download your checklist to keep a copy.
Is the MLOps Roadmap free?
Yes. Reading the MLOps Roadmap and ticking topics is open to everyone. Viewing your plan the first time asks for your name and WhatsApp number once; after that every Uncodemy tool is free and open to you.