A reproducible path
from data to model.
An educational MLOps demo: a logistic-regression classifier trained on synthetic readmission data, orchestrated with Kubeflow Pipelines and tracked in MLflow.
- Platform
- Kubernetes
- Orchestration
- Kubeflow Pipelines
- Tracking
- MLflow
- Data
- Synthetic
Make the experiment inspectable.
A small, connected pipeline for training and tracking.
- 01 / Generate
Synthetic data
Generate a synthetic readmission dataset inside the pipeline.
- 02 / Train
Fit & evaluate
Train and evaluate a scikit-learn classifier.
- 03 / Track
Record the run
Log parameters, metrics, and the model to MLflow.
- 04 / Persist
Keep the artifacts
Store metadata and artifacts on a Kubernetes persistent volume.
Architecture and reproduction instructions: project documentation
Resource-conscious by design.
The project uses standalone Kubeflow Pipelines on local Kubernetes. Its bootstrap can disable optional services on smaller clusters.
Continue exploring
From the model to the system.
Explore more of my engineering work, or discuss a platform and pipeline challenge.