Introduction

Combinator components are intend to be best-of-breed point-solutions for specific ML problems. A combination of components build a stack.

Since there are multiple components that attempt to solve the same problem, components can be organized into catagories. These are:

• Data
• Develop
• Deploy
• Monitor
• Govern

Undoubtedly there are components that cross boundaries. When this happens, they are listed in multiple catagories.

Data

• Minio - S3 API compatible cloud-native storage
• Pachyderm - Data lineage (version control)
• Feast - Feature store
• Istio - Service mesh

Develop

• Pachyderm - Pipelines
• Jupyter - Notebooks
• Kubeflow - Pipelines, Notebooks, Training

Deploy

• KFServing - Serving
• Seldon - Serving and monitoring

Monitor

• Boxkite - Cloud-native model monitoring
• Seldon - Serving and monitoring

Govern

• MLFlow - Model management