Is My Job Title Made Up? Finding Your Role in ML

I remember how stressful it was to search for my first data science job. Having just graduated with my Masters, I spent weeks blindly searching LinkedIn and blanketing job postings with applications like resume_(insert job title here).pdf. Fun fact: I received Greenhouse rejections for jobs I didn’t even know I applied to for months after starting my first job.

While the sheer volume of resumes sent out was most definitely a byproduct of my anxiety, the blanketing of resumes to any job title containing the word “data” was not that far off. There are two scary truths within the world of job searching for data science roles:

1. All companies that know what they’re looking for in a role call it something different

The most common example of this can be seen below. A product analyst at Google is essentially a “data scientist.” While a data scientist at Facebook may be more like a traditional “data analyst.” Companies will follow their own internal guidelines when setting job titles.

Comparison between a recent Data Science job posting at Meta and Data Analyst posting at Google

2. And companies that don’t … call it Data Scientist

Another thing to be wary of when job searching is the vanilla umbrella term of a data scientist. Although your dream data scientist job is probably also titled data scientist, there is a lot of mis-marketing of positions in the ML space right now.

A recent MLOps survey paper called it “headless chicken hiring.” This occurs when organizations know they want to level up their data science and AI capabilities. They need data lakes cleaned, ETL pipelines built, someone to create visualizations and generate reports, and/or any other potential task involving data. And to find people to do it, they utilize the same approach as me, unemployed on LinkedIn in my parent’s basement, just throwing a title with “data” in the name and thinking, “Well, good enough.”

Taking a “well good enough” titled position when looking for a traditional data scientist role leaves a lot of developers unhappy. It’s one of the reasons that data scientist roles have a turnover rate of 55% higher than already high-tech industry standards. So, it is essential to find a position that matches your expectations.

That is why, my number one piece of advice for people who know what they are looking for in their job search is to read the job description

Bonus points for those that not only evaluate the job description to understand what they are looking for but also tailor their resume to match the wording of the posting to optimize for resume screening algorithms. (Google Chrome extensions like Jobalytics will automatically compare your resume to job postings and give you a percent match).

But … that still leaves those who don’t know exactly what they’re looking for. And for them, my second piece of advice is to …

Start by understanding traditional data roles by skills, toolkits, and role in the ML lifecycle

When looking at Google’s job posting, you may have noticed that I compared their role to a “traditional data scientist.” As we’ve established, there are quite a few job titles with data in the name. When starting a job search within AI, it’s essential to understand what the traditional breakdown of those roles is:

Quick drawing covering the different Tasks, Skills, and Toolkits needed for each of the major technical roles in ML

It can be helpful to choose a role you would like to go after and then tailor your resume and/or up-skill to fit into it. You can start with some of the skills/toolkits listed above and cross-check with LinkedIn job postings to ensure they are current.

But, all these skills are so different? How are these all AI roles?

Another thing to think about when applying to roles in AI is understanding where the position you are applying to fits into the ML ecosystem. While I am sure I will cover the trend of the “end-to-end” data scientist in a future newsletter, most organizations still operate with a project “handoff” system. All the roles listed above work together to create ML projects.

Understanding where different roles sit in this system can help job seekers think about the big picture of their future roles:

Quick drawing following the ML Lifecycle

In this image, we can see these handoffs in action. The data engineer takes raw data from the “real world” — i.e., any external data source in this case — and transforms it into usable formats. They maintain the data infrastructure, creating data lakes and, in more recent years, feature stores for data scientists/analysts to pull from.

Next, the analysis and experimentation pipeline begins, as organized data is taken from the data lakes or feature stores. Data analysts/scientists transform that data for business insights/outcomes. Data analysts typically focus more on the insights part of that equation. When external stakeholders ask business questions, data analysts transform and analyze the data to provide answers. These answers could be in the form of charts, reports, or even presentations.

On the other hand, data scientists’ experimentation pipeline also starts with data analysis. However, for them, this analysis is rarely driven by answering a question; instead, they are asked to solve a problem. For example, instead of answering the question, “How has the rate of credit card fraud changed from last year to now,” they are asked to improve the system to detect credit card fraud. To do this, they often work with data analysts to explore data related to the problem and use their statistical background and domain knowledge to evaluate solutions. These solutions are often statistical or machine learning models.

The next handoff occurs after the model is created and ready for production (to be served to end users). While ML engineers often have some hand in model experimentation, they are a crucial part of this handoff, taking the finished model from the data scientists. Solutions created after the analysis/experimentation pipeline are part of the productionalized pipeline.

ML Engineers are in charge of helping evaluate data scientists’ models to ensure they’re ready for production. They then optimize the code, through techniques like data parallelization and the model, through methods like reducing model size. Finally, they own preparation for model deployment from setting up the infrastructure, testing production environments, validating CI/CD pipelines, and maintaining that deployment through monitoring.

After ML Engineers deploy their model, they provide backend engineers or developers with an API definition that defines how to interact with their deployed model. With that API, backend developers integrate it into existing backend architecture and infrastructure. They also often assist ML engineers with monitoring and maintaining the ML model, helping with this like debugging data processing or implementing error handling.

That’s the final hand-off, as the ML model begins to run in production. At this point, it begins making predictions on data. That data, as well as those predictions, are typically stored in the organization’s databases. And the cycle continues …

All in all, if you’re looking for a technical role building out AI systems, it’s definitely attainable. You just need to know where to start and what skills to build up. While we’ll definitely cover some more specific skills in upcoming newsletters, some great places to start are Coursera, Udemy, and even Youtube (the world’s largest online education platform).

Also, it’s important to note that these aren’t the only roles available in AI. Teams needs the PMs (product, project and program managers), technical writers, designers and much more to develop their AI systems and products!