MLOps learning roadmap
Take the model you built in Data Science to production: test it, ship it, monitor it, and improve it safely.
Complete Data Science first
Start with short learning sessions here. Continue coding and project work on your laptop with the same GitHub account.
Start this pathWhat you’ll build
Credit risk ML platform
Your Data Science model, run like a production system on Docker Compose: tracked training, CI gates, safe releases, rollback and monitoring while labels arrive late, built locally and then run on AWS. It continues in your Data Science repo, so the model and its production setup live together.
Tap a milestone to see what you’ll build on your laptop.
1Tracked, Repeatable Training
Bring in your Data Science model, add serving and rollback to the design doc, track runs in MLflow and rebuild it the same way in Docker.
What to show: docker-compose.yml, the training flow, MLflow runs and a data contract.
2CI/CD for ML
Add CI tests that block a release when quality, cost or speed gets worse.
What to show: The CI workflow, regression tests and docs/release_gates.md.
3Serving and Safe Releases
Run a canary locally with Docker Compose, then for real on AWS Lambda with a weighted alias, and roll back when it fails.
What to show: Compose services, routing tests, load results and a rollback that works.
Tools: Grafana, Retrain.
What you'll build
Credit risk ML platform
Your Data Science model, run like a production system on Docker Compose: tracked training, CI gates, safe releases, rollback and monitoring while labels arrive late, built locally and then run on AWS. It continues in your Data Science repo, so the model and its production setup live together.
Data
Your Data Science model and the same UCI credit-card data. It is one historical snapshot, so late labels and shifted data are simulated. Say so wherever you report them.
One GitHub repo for the whole path. You connect it once, and each module adds a verified milestone.
You need: A laptop with 8 GB of RAM, Docker, about 10 GB of free disk, and an AWS account on the Free plan. If AWS will not accept your card, a GitHub-only route gets you through the same checks.
Tap a milestone to see what you’ll build on your laptop.
1Tracked, Repeatable Training
Bring in your Data Science model, add serving and rollback to the design doc, track runs in MLflow and rebuild it the same way in Docker.
What to show: docker-compose.yml, the training flow, MLflow runs and a data contract.
2CI/CD for ML
Add CI tests that block a release when quality, cost or speed gets worse.
What to show: The CI workflow, regression tests and docs/release_gates.md.
3Serving and Safe Releases
Run a canary locally with Docker Compose, then for real on AWS Lambda with a weighted alias, and roll back when it fails.
What to show: Compose services, routing tests, load results and a rollback that works.
Tools: Grafana, Retrain.
A new model passes its tests, then its speed and risk-flag rate drift before the labels arrive. Show how you spot it, roll back, evaluate later and approve a fix.
Module by module
- 01
1Tracked, Repeatable Training
Log every training run, chain the steps into a pipeline and package it so it reruns the same anywhere.
Bring in your Data Science model, add serving and rollback to the design doc, track runs in MLflow and rebuild it the same way in Docker.
Tracking runs with MLflow · Training pipelines · Repeatable runs in Docker
24hTracked, Repeatable Training
Bring in your Data Science model, add serving and rollback to the design doc, track runs in MLflow and rebuild it the same way in Docker.
- 02
2CI/CD for ML
Test every change to code, data and model automatically before it can be released.
Add CI tests that block a release when quality, cost or speed gets worse.
Automated ML tests · Versioned model builds · Approved releases
22hCI/CD for ML
Add CI tests that block a release when quality, cost or speed gets worse.
- 03
3Serving and Safe Releases
Serve predictions in batches or live, send a new model a small share of traffic first, locally and then on AWS, and roll back fast.
Run a canary locally with Docker Compose, then for real on AWS Lambda with a weighted alias, and roll back when it fails.
Batch and live serving · Canary releases · Health checks and rollback
29hServing and Safe Releases
Run a canary locally with Docker Compose, then for real on AWS Lambda with a weighted alias, and roll back when it fails.
- 04
4Monitoring and Incidents
Watch the service, data, model and business numbers, and know who acts and how when something breaks.
Monitor service, data, model and business numbers while labels arrive late, and run an incident drill.
What to monitor · Detecting drift · Handling incidents
26hMonitoring and Incidents
Monitor service, data, model and business numbers while labels arrive late, and run an incident drill.
- 05
5MLOps Capstone
Ship a platform that retrains your model, promotes it only with approval and meets clear service targets.
Release the platform with triggered retraining, approved promotion in MLflow, the champion live on AWS, service targets and a walkthrough video.
Automatic retraining · Approved model promotion · Service targets (SLOs)
30hMLOps Capstone
Release the platform with triggered retraining, approved promotion in MLflow, the champion live on AWS, service targets and a walkthrough video.