DevOps is a software engineering methodology focused on helping software developers (dev) and operations (ops) teams by facilitating a culture of collaboration and shared responsibility. Using iterative software development and automation, DevOps improves productivity across various teams of your organization - helping teams manage the software development lifecycle and drive business outcomes faster.
With the rise of Machine Learning and Artificial Intelligence, DevOps solutions are missing key capabilities focused on ML/AI workloads and the unique challenges of integrating data science into modern web applications. ModelOps aims to bridge this gap.
ModelOps enables your company to integrate cross-functional stakeholders to make sure that your data science workloads (e.g. ML/AI models) are managed, operationalized, and refined in a continuous manner. ModelOps tools allows you to plan, build, test, deploy, secure and observe ML and AI workloads.
ModelOps itself is not one feature but instead a conglomeration of a wide range of integrated features across traditional devops systems that allows you to streamline workflows, reduce complexity, and speed up development that includes:
Native data connectors - Turnkey connectors for major data warehouse providers to provide easy connectivity between data stores and ML/AI workloads
ELT engine - Ability to extract, load, transform data from disparate data sources and get it into the right shape/form for processing.
Code Review Workflows - Set code review policies to govern how code and models can be reviewed and approved before being deployed
Feature Store/Data Catalog - Reduce repetitive data processing and curation work required to convert raw data into features for training an ML algorithm
Security & Compliance Policies - Manage and enforce policies for code and models to minimize security risks and promote adherence to compliance rules.
Data versioning - Most data is not static, it’s dynamic, versioning is focused on tracking and storing changes to data schemas, lineage and behavior over time.
Enterprise Python notebook experience - Rendering, visual diffing, code review, executable cells, dependency management, environment management, compute connectors as well as discovery and explorability.
Agile Planning - Manage and plan your projects and programs within your team or across your enterprise with integrated Agile support.
Specialized Compute - Powerful cloud machines with great CPU/GPU/Memory allocations appropriate for ML/AI workloads
Workload orchestration - Build, test and deploy ML models based on task sequencing that you specify. Inclusion of advanced CI/CD features can help reduce wasted compute like auto-stops, job caching and checkpoints.
Security Scanning - Run security scans of your code and models to identify, track and resolve security vulnerabilities
Model registry - Version control for machine learning workloads. Store, discover and verify model artifacts and model metadata.
Model observability - Inspection into models and visualizations to understand how and why ML models produce the results they do. Data check, performance analysis, monitoring and explainability.