Name: Lorenzo Ostano
Title: Software engineer, VX Fiber
Location: Milan, Italy
Education: B.S. and M.S. Industrial Engineering, Politecnico di Milano
Favorite Machine Learning Area: Natural Language Processing
Lorenzo Ostano first encountered machine learning and data science as a business analyst shortly after graduating from college. He later took the Deep Learning Specialization, and soon after landed work as an engineer. After working several years for a variety of consulting companies, he recently took a job as a software engineer. He spoke to us about why he believes traditional computing skills are important to deploying enterprise machine learning applications.
You recently started a new job as a software engineer. Why did you decide to pursue a non-machine learning position?
LO: Yes, I am now a software developer for an internet service provider based in Sweden called VX Fiber. At the moment there is no machine learning at this company, but I’m sure there will be in the future because, as an internet service provider, we have loads of data. I’m still learning a lot, but I’m pretty sure there are opportunities for automation.
My new role has no machine learning in it because I want to get experience developing enterprise applications from the back end. For the work I want to do, which is machine learning, I think the biggest foundation is software engineering. I’ve seen a lot of data science experiments that remain in Jupyter Notebooks. Without software engineering, you wind up with pet projects that don’t ever make it to the market.
How did you first get interested in AI?
LO: The first time I heard about machine learning was 2017. I was just getting into data analysis and saw that what you could do with machine learning was revolutionary. And it seemed complicated, but it also seemed like something I could do.
What was your career like before you got into machine learning?
LO: I graduated in 2012 with an undergraduate degree in industrial engineering. Then I moved to Australia where I did a lot of odd jobs. My first entry into basic programming was as a business analyst. That’s also where I found out about data science and machine learning.
How difficult was it for you to get your first AI job?
LO: My first data science job was at a French consulting company called Alten, starting in late 2019 after completing the Deep Learning Specialization. The recruiting process wasn’t that hard. I think the director was impressed by my passion for machine learning. He saw that I wasn’t skilled in software engineering but was able to talk in depth about topics like natural language processing, computer vision, and time-series forecasting.
You have both bachelor’s and master’s degrees in industrial engineering. Are there skills from your college experience that are transferable to machine learning?
LO: Yes, definitely. Advanced mathematics like calculus are the foundation of any engineering degree, and those also form the foundation of data science and machine learning.
Prior to your current job, you worked as a machine learning engineer at OmniNext. Can you tell me about your work there?
LO: OmniNext is like a startup container, it has many different projects. When I joined, I was put on a team with two other machine learning engineers. We worked on a variety of things, but I spent about 80 percent of my time developing a product using time-series forecasting for renewable energy. The goal was to make the electricity market more efficient by knowing in advance how much energy would be coming from renewable sources.
We started with just one year of historical data from an Italian energy company. And of course, one year of data is nothing for this type of challenge. Eventually we found more data from a Spanish website that collected data from a government-run wind farm, which led to even more data from a different Italian company. This gave us about four or five years of wind data, which was enough to develop a really robust system for predicting how much electricity would come from the wind. My former colleagues are now trying to sell this product.
How do you characterize the demand for machine learning engineers in Europe?
LO: Demand is pretty big, but I think most of that is driven by the fact that machine learning has become a buzzword for innovative. In my country, at least, many companies do not have the necessary engineering culture to develop machine learning products that are ready to compete on the market. When I was interviewing for work, I spoke to a lot of consulting companies that were pitching machine learning as the go-to solution.
A lot of these companies want to say they do machine learning, but at the end of the day, they probably will not be successful. Some companies are doing great things with the technology. These are businesses with a lot of data, like banks and financial institutions. I also see a lot of consulting companies looking to hire machine learning engineers or data scientists. I think the demand for machine learning from manufacturing is currently underdeveloped, and there is a lot of opportunity to grow in that space.
Do you have any tips, lessons, or advice to pass along to other people who are starting to look for their first machine learning job?
LO: One tip is to build a portfolio of projects in a wide range of topics. When doing projects, you are going to learn more if you can find problems that you really want to solve. Then you want to post about your work. The simplest way is to post your work on Github, but even better would be to write a blog about your experience.
If you want to get into bigger companies, you need to study up on various computer science topics. I tried to interview as a machine learning engineer with several consulting companies and a few high-growth international start-ups before I landed my current job. With the consulting firms, the technical questions were pretty hard. I felt like I was getting interviewed for multiple positions at once, from full-stack developer to data engineer. These interviews also included a good dose of weird brain teasers and questions about data structures and different algorithms
What were you least prepared for when it came to doing machine learning in the real world?
LO: Definitely the software engineering aspect of deploying machine learning products. Another thing I recently learned was the importance of domain knowledge. For example, if you’re doing work on renewable energy, you should connect with a stakeholder who really knows the business so you can understand their requirements and priorities. At the end of the day you are trying to solve business problems, not just computer science problems.
How do you stay up-to-date on the latest machine learning research?
LO: I read your newsletter, The Batch, because it has all of the most relevant stuff in a condensed format. It makes it really easy to skim and keep up with the general trends in AI. For deeper dives, I follow several YouTube creators, like Sentdex,Machine Learning Mastery, andPyImage Search — which is great for its focus on computer vision. I also follow many passionate researchers and engineers on Twitter.
You can find Lorenzo Ostano onLinkedIn.
Ready to #BeADeepLearner like Lorenzo?