# Maturity Levels and Roles — MLOps

Source: https://www.skillbyai.com/en/mlops/i-maturity

> Grow automation step by step.

## Manual, pipeline, continuous

Teams typically move through levels: **manual** (notebooks, hand-copied model files, ad-hoc deploys); **automated training pipelines** (reproducible training, tracked experiments, a registry); and **continuous training and delivery** (pipelines triggered by new data or schedules, automated tests and gates, monitored rollouts). Roles overlap: data scientists build models, ML engineers productionise them, data engineers own pipelines, platform teams provide shared tooling. Aim for the level your number of models and their risk justify, not the most sophisticated stack.

## What each level looks like

Use it to plan the next step.

```text
level 0 manual       notebook -> model.pkl emailed -> manual deploy; no tracking
level 1 pipelines    scripted training, experiment tracking, registry, repeatable deploy
level 2 continuous   triggers on new data, CI tests + evaluation gates, canary rollout,
                     monitoring with alerts, automated retraining
```

## Match tooling to scale

Two models updated quarterly do not need a full platform; a tracked script, a registry and monitoring may be enough.

**Quiz:** What characterises level 0 (manual) MLOps?

- [ ] A model registry with gates
- [ ] Fully automated retraining
- [ ] Canary releases with alerts
- [x] Notebooks and hand-copied model files with no tracking

*Answer:* Notebooks and hand-copied model files with no tracking. Manual processes are fragile and hard to reproduce.