About This Constellation ShortList™
With the explosion of artificial intelligence and ML models in the enterprise space, productionizing ML models is very expensive for many digital-native companies. The solutions for MLOps are expected to become a $4 billion market by 2025. Data scientists are not full-stack programmers, as their skill set is limited to experimenting and developing the best model to solve a specific business problem. If the model is accepted by the business owners, then deploying it in production in the shortest time possible before competitors do can provide a distinct business advantage. The plethora of MLOps frameworks, both commercial and open source; a fragmented ecosystem; and fragmented data sources make this task harder for deployment teams. Integrating all these ML capabilities, tools, platforms, and practices with existing application lifecycle management systems is extremely difficult and time-consuming.
The art of model development, model training, model deployment, model monitoring, and model governance involves major engineering work that is generally managed by data engineering teams. This is the area where development operations (DevOps) practices are applied to ML. It involves creating production-grade models, deploying them, and refreshing them as quickly as possible. It means automating the majority of these tasks so that the models can be deployed with one push of a button as soon as they are ready and approved by the stakeholders.
Constellation considers the following criteria for these solutions:
- Support for multicloud environments. The chosen MLOps tool/platform should support—at a minimum—AWS, GCP, and Azure.
- Its ability to allow and manage the deployment of ML models on Kubernetes with ease.
- Connectivity to as many commonly available data sources as possible.
- Integration with major continuous integration/ continuous delivery (CI/CD) and DevOps tools.
- Support for the following model capabilities:
- Model versioning
- Model training & experimentation
- Model deployment
- Model validation
- Model monitoring
- Model retraining
- Model registry
- Feature engineering
- Model serving
- Model governance
- ML pipeline automation
The Constellation ShortList™
Constellation evaluates more than 20 solutions categorized in this market. This Constellation ShortList is determined by client inquiries, partner conversations, customer references, vendor selection projects, market share and internal research. Of those solutions, the following are short-listed to be in our future market overview deep-dive MLOps report.
- AWS SAGEMAKER
- DOMINO DATA LAB
- GOOGLE VERTEX AI
- INTEL CNVRG
- MICROSOFT AZURE
Frequency of Evaluation
Each Constellation ShortList is updated at least once per year. Updates may occur after six months if deemed necessary.
Constellation clients can work with the analyst and the research team to conduct a more thorough discussion of this ShortList. Constellation can also provide guidance in vendor selection and contract negotiation.