To cover model development, training and verification, ENFINT MLOps Platform provides managed development environments based on Jupyter, Visual Studio Code and RStudio.
To track experiments and register models, integrated MLflow component is ready.
It also provides a web interface allowing users to track logged results and catalog both candidates and production models using MLflow Model Registry.
To orchestrate batch jobs Apache Airflow along with Git integration is used.
The component handles not only model verification but also deploying of complex batch models. Such models may rely on ETL processes, industry-proven frameworks (e.g., Apache Spark) and variety of other solutions.
To run models as online services, Seldon Core component provides quick deployment pipelines.
Target REST service can be deployed with no extra coding, based on previously registered model in MLflow. As an advanced option, an inference graph can be customized additionally to support input and output transformers, ensemble of models or even A\B testing.
To discover, manage and serve your data, ENFINT MLOps Platform has built-in feature store component based on Feast and its API. It allows users to classify and manage data properly within historical storages and online databases.
Feast web interface is helping users by proper visualization of configured features, feature groups and services.
To cover monitoring part of the lifecycle, ENFINT MLOps Platform provides a set of monitoring capabilities based on Evidently, Grafana, Prometheus and Jaeger.
The combination allows to monitor your models continuously from different angles, analyzing tracing data, technical metrics thresholds and various business data through a set of available monitors. Triggers can be configured to notify users when certain conditions are met.