Machine Learning Interviews

Chapter 6. Technical Interview: Model Deployment and End-to-End ML

In Chapters 3 and 4, you got an overview of important interview concepts related to ML algorithms, model training, and evaluation. For ML models to have an impact on users, whether they are your company’s customers or internal users and fellow employees, the model needs to be deployed.

There are many levels of being deployed, but the most important thing is that the end goal for the model is achieved. If your model is being run manually and ad hoc every time the marketing team asks for new results, and that’s working well, then that could be your level of deployment. Or, you could have a fully automated system where the model goes out to customers as part of an A/B test, without the person who trained the model needing to do anything beyond the model training. That could be the level of deployment in a situation where the end goal calls for it.

On that note, it’s not required for ML professionals to know all the details of model deployment. If you’re applying for one of the following jobs, however, it would be useful to brush up on the topics mentioned in this section. Roles that are likely to require deeper knowledge on model deployment include:

• Machine learning engineer that doesn’t only do model training

• MLOps engineer

• Data scientist or MLE at a startup that doesn’t have additional dedicated people working on ML deployment

Coincidentally, the more your job falls into this type of role, the higher the possibility ...