Get startedGet started for free

Laying the foundation for MLOps

1. Laying the foundation for MLOps

We discussed that MLOps is not a narrowly defined standard but a loose set of concepts, tools, and best practices. How can businesses start and foster their MLOps journey? This will be the topic of this video.

2. Agenda

Here, we will focus on the required skills, technology, and culture. Let's start with the skills we need!

3. Skills required for MLOps

MLOps requires different sets of skills. First, the best infrastructure will not help us much without the right people. We have seen before that MLOps lies at the intersection of machine learning, data, and software engineering. Critical roles include, first of all, a business or subject matter expert who needs to provide the business question to be addressed through MLOps. Second, a data engineer is essential to provide high-quality data for the machine learning application and to manage the entire data life-cycle from the original data source to the final prediction. A data engineer will work closely with both the business expert and the data scientist or machine learning engineer. Third, the tasks of a data scientist or machine learning engineer include developing, testing, and verifying the models, as well as communicating the necessary technical information to the business. Fourth, we need a software engineer who ensures that the machine learning application fits neatly into larger IT systems. Those tasks also include adherence to company-wide code guidelines, among others. Lastly, there is also the role of a solution or machine learning architect.

4. MLOps architecture

This role is responsible for designing the overall technical infrastructure and defining the tools used. Do not take this job role lightly: modeling constitutes only a small part of MLOps. Scaling machine learning requires a reliable, efficient, and highly automated technical environment. The graph shows you what a prototypical MLOps architecture can look like. The box sizes roughly correspond to the amount of code involved. A lot of code will go, for example, into configuring the infrastructure or collecting the data.

5. Where to host your MLOps platform

You do not need to create such an infrastructure from scratch. Some companies offer an end-to-end platform to host and operate your machine learning applications. These include, among others, the big cloud providers AWS, Azure, and GCP. Other companies like, for example, Dataiku or Datarobot, also provide fully-managed end-to-end platforms. Then, you can also build up your tool stack on your own. You can also combine these options: building your custom tool stack on, for example, AWS to easily move to another cloud provider and avoid cloud-provider lock-in risks.

6. Requirements of MLOps platforms

To achieve scalability, we need to ensure that we do not develop each machine learning model and each application from nothing. This requires common standards and reusable components in combination with highly-automated workflows and continuous testing and validation.

7. Cultural aspects of MLOps

Culture is as important as people and technology. How can we foster the streamlining of MLOps? Here are a few things businesses can do. The upper management should give clear strategic guidance about the business objectives. There need to be close cooperation and communication between all relevant teams, such as IT, data, and business. You can ensure this via joined responsibilities. Essential is also a culture of continuous, incremental improvements and delivery, combined with autonomy and regular "blameless" feedback. Finally, we also need relevant skill investments for everyone involved, including, for example, the end users of MLOps applications.

8. How to get there

How can a business become MLOps-ready? First, you need to start with a clear business goal. Second, you will fail if you cannot get and retain the right talent or do not provide incentives to collaborate. Last but not least, think carefully about a well-designed central architecture.

9. Let's practice!

We will discuss all this in more detail in the following chapter but first, let's practice!

Create Your Free Account

or

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.