I am the first Machine Learning Engineer hired in our Data Science team. I just started working in this role, so take my comment with a grain of salt.

Much of my job helps fill a particular niche that many people in my team would rather not focus on or they are not skilled yet to do.

I collaborate closely with our Data Scientists and often read, reinterpret, and transform what they wrote as a Notebook or a Python script into software that can be deployed. This could mean parallelizing, optimizing, tuning, testing, and wrapping their code that exists as a proof of concept into something that can be deployed.

I think because Data Science is a relatively new career path and that people come from various different backgrounds, defining what is ‘Data Scientist’ is particularly challenging. The responsibilities of the Data Scientist demand a very broad skillset and that’s why we are seeing more roles and titles such as “Machine Learning Engineer”, “Analyst”, “A/B Testing expert” to help clarify exactly what they focus on. For instance, while I know a lot about programming languages, algorithms, machine learning, classification, neural networks, and numerical methods, I don’t know much at all about economics, operations research, and experiment design.

Admittedly, many of the Data Scientists that I work with have hardly any professional programming experience at all except for working with R, Stata, or SASS, yet they are very skilled in mathematics and modeling. Many have PhDs or Masters in Statistics, Operations Research, Economics, or Neuroscience. Many have not been exposed yet to using version control and pull requests to collaborate on large scale software projects. They are, however, amazing modelers and know a tremendous amount of mathematics.

I was told that why I was a competitive candidate for the Machine Learning Engineer position is that I recently came from a Software Engineer position where I was a team lead and ScrumMaster. My experience in Agile, Scrum, writing test cases, CI/CD (Jenkins), DevOps (Docker+Kubernetes), etc- were seen as a benefit to help make our Data Science team more effective in the organization. I completed my PhD in Computer Science which had a strong applied Machine Learning direction and that certainly helps with communication and collaboration with the data science team.

I generally take a more computer science approach to the problems. For instance, how can we parallelize this task? How can we make this component more modular and reusable? Can we profile this application and employ optimizations to improve the training or classification time? What kind of algorithms or data structures can we employ? How can we create developer tools to simplify tedious tasks?

I would say that a successful Data Science team should have people from all sorts of backgrounds so that everyone can learn from one another.