Reference architecture: Fully automated MLOps

Fully Automated MLOps

Arturo Opsetmoen Amador

Senior Consultant - Machine Learning

What is a reference architecture?

A reference architecture is a template we can use to design solutions in different IT domains.

A set of documents with:

Structures and integrations of IT elements
Patterns commonly present in IT systems, including ML systems
Leverage experience and best practice from industry players

Fully automated MLOps architecture

An image showing a high level overview of the fully automated MLOps reference architecture.

¹ https://cloud.google.com/architecture/mlops-continuous-delivery-and-automation-pipelines-in-machine-learning

Reference architecture - Orchestrated experiments

An image showing a high level overview of the fully automated MLOps reference architecture. The image zooms in to a set of connected boxes. From left to right, data analysis, orchestrated experiment, source code, source code repository.

Reference architecture - Source code & CI

The image is shifted to the left. After the source code repository, a continuous integration block is showed. This block has a build box that produces packages as output.

Reference architecture - Artifacts & CD

The image shifts down to show a continuous deliver box.

Reference architecture - ML pipeline deployment

The image is shifted to the left to show an automated ML pipeline with different steps: data extraction, data validation, data preparation, model training, model evaluation, model validation.

Reference architecture - The metadata store

The image shifts down to show the Metadata Store collecting logs generated by the ML pipeline.

Reference architecture - The model registry

The image shifts up and to the right to show that models trained by the ml pipeline are pushed to a model registry.

Reference architecture - Prediction services

The image shifts down to show that models trained by the ml pipeline are continuously delivered to a prediction service module in the architecture.

Reference architecture - Continuous monitoring

The image shifts left to show that there is a performance monitoring component that tracks the performance of the predictions delivered by the prediction service.

Reference architecture - Automated trigger

The image shifts further to the left to show that there is a trigger component connected to the performance monitoring module

Reference architecture - Automated retraining

The image shifts up to show that the trigger can activate the automated ml pipeline in the system.