MLOps accelerating the adaptation of AI in enterprises - three concrete cases

Machine Learning Operations, or MLOps, is a set of systems, tools, and practices that enables the deployment and operation of machine learning pipelines in production in a repeatable and trusted manner. MLOps accelerates the adaptation of AI in enterprises by introducing common tools and processes to organize the work across multiple teams, organizations, and lifecycle phases. MLOps achieves this by looking at machine learning solutions from an end-to-end perspective - from the data collection phase to monitoring and back, tracking how business results are getting achieved and what to do with the models and data to make further improvements.

In this video, our Chief Technology Officer, Niko Vuokko, introduces the topic by going through three concrete cases we at Silo AI have been working on.

• 01:37 Case: Global Retail Chain
• 02:39 Case: Data Service Provider
• 03:23 Case: A Global System and Software Vendor

As with any new technology, the real benefits of AI do not come from the technology itself but from how the organization’s ways of working can be improved. Therefore individual AI development programs should be seen also for their educational value, helping the organization move towards scaling AI. Here the purpose of MLOps is to provide an efficient technical framework around which also human processes in development, operations, and governance can be evolved.

From a business perspective, MLOps is the bridge between the experimentation world and the production world: building MLOps means moving beyond the first AI pilots towards a more mature, operationalized, and scalable way of working with AI.

Helped a global retail chain to scale AI usage across regions and use cases

In the first case example (01:37 Case: Global Retail Chain), we worked with a global retail chain to enable the data science teams to scale their AI use across their e-commerce store with a combination of components from open source technologies and from the Azure cloud environment. Before Silo AI, the company had already identified and implemented several tens of use cases for AI. However, the models were built on data science environments isolated from end products. We’ve worked in deep collaboration with the client team to define, roadmap, and build a platform for model development, deployment, and monitoring that helps all the AI product teams to deliver AI-driven e-commerce features connected to online systems. The platform allows also efficiently sharing each team’s work across the organization to enable cost-efficient scaling through common model components.

From manual machine learning workflows to automated, scalable way-of-working

As another example (02:39 Case: Data Service Provider), a data service provider had well-defined MLOps requirements but needed help in making technology choices and building a solution that would meet their needs for scalable and rapid machine learning model training and reporting. Silo AI experts built a model training infrastructure and automation to speed up their ML workflows using open source tooling and an AWS cloud environment. This setup also created new capabilities for our customer to collaborate more deeply with its own clients, allowing for much better possibilities to accommodate client-specific requirements in the development work. 

From MLOps exploration to an integrated team for global systems and software vendor

As a third example (03:23 Case: A Global System and Software Vendor), we started with our client, a global systems and software vendor, by defining the requirements and technology options for a sensitive private cloud environment containing, among other things, medical data. The outcome was a customized and standardized approach for machine learning model training automation that accelerates both consistent machine learning experimentation in concept development work and also AI feature delivery to the final end products. In addition to this, the standardized machine learning workflow and tooling brought data scientists and DevOps professionals closer together, enabling a more integrated approach to collaboration across different work phases and accelerating the overall pipeline of work from research and concept phases to final product delivery.