MLOps Maturity Model

Question

Hey,

i'm right now trying to understand the MLOps Maturity Model. The thing i don't understand is the Level 1 (DevOps but not MLOps). Why is the data gathering at this stage automatically, when data is unique for MLOps and is no part of DevOps? And the second thing is, why are the different people not working together at that stage? The DevOps principles say, that the Development and the Operations are working together and communicate with eachother. And in the case of MLOps the Developers are the Data Scientists and the Data Engineers. So why are they siloed from the Software Engineer (Operations)?

Thanks in advance

Answer 1

Hello @MacerPacer

Thanks for reaching out to us, below answer is based on some researches I got across the internet, please let me know if you have any point you feel confused.

For your question 1, why is the data gathering at this stage automatically, when data is unique for MLOps and is no part of DevOps?

Level 1 representing organizations that have adopted DevOps practices but have not yet integrated MLOps practices into their workflows. Regarding the data gathering aspect of Level 1, it is important to note that data is a critical component of both DevOps and MLOps. In DevOps, data is often used for testing and monitoring applications in production, while in MLOps, data is used to train, validate, and test machine learning models. Therefore, it makes sense for organizations at Level 1 to have some data gathering processes in place, even if they have not yet fully integrated MLOps practices into their workflows.

For your question 2, why are the different people not working together at that stage? The DevOps principles say, that the Development and the Operations are working together and communicate with eachother.

Regarding the siloed nature of different teams at Level 1, it is true that DevOps emphasizes collaboration and communication between development and operations teams. However, in the context of MLOps, there may be additional teams involved, such as data science and data engineering teams, that are responsible for building and managing machine learning models. These teams may have different expertise and tooling requirements than traditional software engineering teams, which can lead to silos. However, as organizations progress through the maturity model, they can work towards breaking down these silos and fostering greater collaboration between different teams involved in the MLOps process.

For your question 3, in the case of MLOps the Developers are the Data Scientists and the Data Engineers. So why are they siloed from the Software Engineer (Operations)?

We have explained some parts of it in the Q2, it is mainly because of the different tooling and process as below.

In the case of MLOps, it is true that developers, specifically data scientists and data engineers, are responsible for building and managing machine learning models. However, software engineering teams, specifically operations teams, play a critical role in deploying and managing these models in production environments.

The reason for the potential silos between these teams is that they may have different areas of expertise and use different tools and technologies. For example, data scientists and data engineers may be more familiar with tools such as Jupyter notebooks and data processing frameworks, while operations teams may be more familiar with infrastructure automation tools such as Kubernetes or other approaches.

I hope my answer helps, please take a look and let me know if you have any other question. Happy to help further. Thanks a lot.

Regards,

Yutong

-Please kindly accept the answer if you feel helpful to support the community, thanks a lot.