Overview
Verta is a complete MLOps (ML Operations) platform focused on operationalization of ML models, i.e., integrating ML development and delivery into regular software in a way that allows data scientists to continue focus on machine learning and data science, while providing ML and DevOps Engineers the means to safely and reliably integrate ML into the broader software ecosystem in any organization.
As shown below, the full ML lifecycle consists of three components: the data preparation loop (including ETL, Data Cleaning), the model development loop (including training, feature pre-processing, model validation), and model operationalization (including packaging, release, monitoring and operations). Verta comes in during model development and provides model versioning capabilities via ModelDB. ModelDB then serves as the connection between the development and operationalization phases. During the operationalization phase, Verta provides modular components for model deployment, release, monitoring, and maintenance.
Verta is an open-core platform; i.e., the platform is based on core open-source technology developed by the Verta team that is freely available. Find more information about our open-source technology here and in subsequent sections.
Verta provides MLOps functionality in three key areas: model versioning and metadata, model deployment and release, and real-time model monitoring.
The first step to enable operationalization of ML models is to make them reproducible and associate governance data with them. Verta's open-source ModelDB component is the only system to provide full model versioning and reproducibility along with a rich metadata system.
Head over to overview/versioning to learn more about to use ModelDB in your ML workflows.
The most challenging and yet crucial operation in operationalization of models is model deployment and release. Due to the diversity of ML frameworks and libraries, along with the disparate systems for ML development vs. software delivery systems, it takes many months to release models into products.
One of Verta's key innovations is our open-core model deployment and release system that works seamlessly with models built in over a dozen frameworks and languages, and integrates with state-of-the-art DevOps and software delivery systems.
Head over to overview/deployment to deploy your models.
Once a model is deployed, close monitoring is required, both at the systems level (e.g. CPU, memory, network) as well as the feature and data level to ensure that the model is performing at the highest levels and rapidly remedy production incidents.
Verta's model monitoring provides both system and data monitoring to ensure real-time model health. Head over to overview/monitoring to learn more about how to monitor your live models.