Fantastic ML Engineers and where to find them

New titles in the Data/AI sphere are like cherries. Each year you see some new ones appearing and they will last at most a season. But sometimes the rise of new needs, new business objectives will make one of them stand out.

This is the case of the ML engineer. A strange creature akin to a centaurus : feet of data scientist, head of a data engineer. Data has this tendency to create mythical beasts in its ever-changing history.

As for any trend there will be many articles detailing the whos and the whys. My goal is to share my own experience with a ML engineer profile in a data team in a real environment.

Don’t tell me the sheep with 5 legs doesn’t exist

This was one of my main disenchantment growing up as a manager. The data scientist who can analyse the business case, create the machine learning model, put it in production, pitch it to the Executive committee and then clock out for a hardly earned beer with his team (what a strange sentence in a covid time) exists. But he is rarely available or will burn out very shortly.

Being a firm believer that my team should sleep more than 5 hours every day I had to find a solution. Transitioning from a POC driven approach to a Product driven was not an easy journey. ML products never behave like you would predict, being fundamentally stochastic. Mix it with a pretty capricious data scheduling software (Hi Airflow how you doin ?) and you get a chaotic production environment.

So we needed an expert. Someone that could tame a cruel XGBoost Classifier while still being able to read complex logs and deploy an infrastructure in a jiffy. I re-read Fantastic Beasts and Where to Find Them and was ready to go. Now that I have found several here’s my magical recipe :

As a perfect hybrid, ML engineers should master many tools. Here’s an example of a technological stack based on Google Cloud Platform :

From discussing with data engineers on the Kafka topic feeding for your real time grandma anniversary prediction to cleaning this commit you made on the wrong branch the ML engineer is the firefighter of your data team (datafighter ?). They should be prepared to face hardships, challenges and many dirty iteration of your favorite embedding algorithm.

Being that you want to infuse a dev-ops culture to your ml team or a ml backgroung to your engineering team he/she will fill a central role that will allow everyone to work on the tasks they are the best at.

How many circles will we need in the long run ?

Data Science and ML will be completely integrated in many companies’ strategy. The recent crisis has enabled companies to leverage AI for critical decisions and mitigate losses. This will come with increased accountability needs.

MLOps Platforms will become the norm, based on a complete Data Governance Strategy that encompasses a real product driven approach. In an ever growing customer centric world, you will need the most mature systems to nurture them and get the most out of your relationship. ML Engineer will clearly be key people in order to maintain, develop and innovate on these platforms.

Originally published at https://eacquarone-19999.medium.com on May 7, 2021.

Related posts:

To many computer scientists from the 90’s, AI was something which everyone tried but didn’t really work. But now, AI is rapidly changing the world, changing the way we interact, be it online or…