Workflows

kaos is not functional as an ML platform until it has been deployed in a user-specified cloud environment (i.e. via kaos build deploy). Note that kaos build deploy is distinct from the typical ML workflow in kaos since it should be done infrequently and with greater knowledge of underlying infrastructure.

kaos Personas

The separation is purposely designed to split the intended user flows - build vs. interact. This aligns with the three intended kaos personas.

System Administrator
Superhero
Data Scientist
System Administrator

Deploys kaos infrastructure and generates a running endpoint via kaos build deploy

  • Refer to Build for high level concepts related to kaos deployment

  • Refer to Deploying Infrastructure for low level instructions on deploying kaos infrastructure

Superhero

Deploys and interacts with kaos via kaos build deploy

Superheros have kaos automatically configured - kaos init can be skipped!

Data Scientist

Interacts with kaos core (i.e. experiment, train, serve, etc...) via kaos init

  • Refer to Interact for high level concepts related to kaos usage

  • Refer to Quick Start for low level instructions on how to train and serve a model

Data Scientists require an endpoint from a System Administrator for kaos init