⚙️ CI/CD for Machine Learning

About this simulation

Every push to a machine learning repository can trigger an automated GitHub Actions workflow: a sequence of jobs that lints the code, runs unit and data/model tests, builds a container image, and — if every gate passes — deploys the new build. This scene visualises that pipeline as a glowing track: a commit token travels from stage to stage, each platform lighting up while its job runs and turning red or green when it finishes.

How it works

• Lint & static analysis — fast checks on code style and imports.
• Tests — unit tests and data/model validation; this is the stage most likely to fail, and its failure rate is adjustable.
• Build image — packages the code and model artifact into a container.
• Deploy — the final gate; only a build that cleared every previous stage reaches it.
• If any stage fails, the pipeline halts, the platform flashes red, and the run is recorded as failed — nothing reaches deployment.

Controls

• Pipeline speed — how fast each job runs, purely visual pacing.
• Test failure rate — probability the test stage fails on a given run, like a flaky test suite or a real regression.
• Run lint & tests in parallel — toggles whether GitHub Actions jobs fan out into parallel lanes (faster) or run one after another.
• Auto-push on green — automatically fires a new commit shortly after each run finishes, so you can watch the pipeline cycle continuously.
• 🚀 Push commit — manually triggers a new pipeline run. Drag to rotate the view, scroll to zoom.

Did you know?

Real ML pipelines often add stages this diagram simplifies away — data validation, model evaluation against a held-out set, and canary rollouts that shift only a small slice of traffic to the new model before a full release.