Gabriela Alessio Robles

Please tell us a bit about yourself, and how you got into data analysis.

I'm originally from Mexico, born and raised. I got my undergrad there in chemical engineering, and then worked on obtaining a master's in economics and finance.  I realized that I really wanted to get back into numbers and engineering. I ended up switching and earned my MBA with a concentration in analytics, which got me started in the data science world.  I started at Salesforce, then had a very fun and long period at GitHub where I feel like I grew a ton as a data scientist, followed by a stint at Atlassian. Currently, I'm senior analytics engineer at Netflix, leading several projects.

Why did you choose to study data analysis at MIT Professional Education?

I started the MIT program because I was doing a lot of coding and building, but I really needed to brush up on my technical understanding.  To take a good product to a very solid and robust and almost foolproof product, you need to understand what's happening with the code and the math behind the scenes.

What did the course do for you?

I now enjoy understanding everything that's happening in the code and the math, and being able to communicate that, both to technical stakeholders and also to my team; and also knowing how to explain it clearly and in layman's terms to directors or peers who are not technical themselves.

Is the course applicable to your work in other ways?

I profited from the program, in many dimensions.  It was very well presented. Great materials. Everything that I learned, I was able to apply it in my work. For example, I started making sure that errors were within healthy ranges. By tracking that and making it visible to all stakeholders, we would have a more reliable data source and an economical single source of truth for all the decisions that we were making.

Can you say more about that?

There are many ways of measuring an error for a prediction that you make, or a suggestion if we're talking about Netflix's movies or Spotify's songs. If you're opening up a book or going to stack overflow, probably by default, you'll get the most generalized way of measuring errors, which is RMSE (Root Mean Square Error). But for certain specific cases, that's not the right way to measure errors. And if you're not very solid in how you understand the math behind calculating that prediction or that suggestion, and then the error that comes along with it, then you might just overlook it and say, "Yeah, this looks great. Let's push this out to production." So for me, it was a very useful class.  I went back and I rewrote everything, wrote a little memo to go with it, explaining why this was not the right way to measure that error, why we had to look at it from a different angle. That insight was incredibly useful.

How does this course differ from your prior data analysis learning?

What's great about the course is that they cover so many of these little nuances that, like I said, opening a book or stack--without what they teach in this course, you might not ever get that level of insight. For some of the more deep learning focus courses, it's not the same to predict a song or a movie than, say, a cancer tumor.

Having that context of how relevant it is for you to truly understand your math and then go translate that into code, is what really makes the classes and the program and the MIT faculty that you get access to, very valuable.

Tell us about the capstone project for this course.

I was at Netflix, so I chose the song recommendation engine, because I thought it would be more useful and it has many ways of being tackled. You can do the recommendation by using very little signals and features, or you can go crazy and use the actual audio files and be very, very refined with your recommendations. I picked that as my capstone project because it was very much aligned with what I could be using in the future.

Tell us about the nuts and bolts of taking the course— how did you do that and work full time?

A course can be really hard to fit into a week, but MIT does it super seamlessly and in a very engaging way.  I'm on Pacific time, so the lectures were very early in the morning, sometimes 7:00 a.m., 6:00 a.m., which can be terrible (laughs), but also great because I was able to attend the lectures before starting my day at work. So that was great. I would get the coursework started after the sessions, during the week, after work.  

For the rest, I would commit and sit down during the weekend. When you're going through it, you don't notice the time.  If you like the topics and when you get immersed in the lectures, it doesn't feel like a burden to be spending that much time. It is a commitment, but it's also what you're investing in yourself and your ability to make a difference in your workplace.

Several MIT professors cooperatively teach this course. How did that affect your experience?

Professor Tsitsilkis is so good at going through the concepts. Professor Jegelka covers topics around deep learning around computer vision.  Professors Uhler, Dahleh, and Shah--all of the instructors were amazing. The weekly lectures help to solidify the technical concepts. And they're all applying these concepts on various industry problems.

What topics and concepts were your favorite parts of this course?

I got a very good refresher for some basics that I was already using, but wasn’t being as eloquent when explaining  regression and trees to not-so-technical stakeholders. That part of the course was perfect for me. I almost recorded everything they said, and I was like, "Yes, this is how you explain what a tree does and where it stops and what the branches mean" and so on.

The neural networks sessions were also fascinating and useful, including really good examples.

How did the coursework add value ?

Over the weekends, with your coursework, the lectures come alive when you start coding but you understand what you're doing. Because, if you only code and deploy and it works, you won't learn as much as when you know exactly what's happening with a library that you're using, or if you're hard coding things and just go through it on your own, which is what I did, I wanted to go back to basics and try to be as simplistic and minimalist as possible, then it starts clicking in.  

Is data science be something that someone can use in any company, in any industry?

There's data science in everything. You can apply it to the tiniest problem or the largest one. Recommending a song might seem trivial, but there's actually a lot of personalization and different ways that you can go about it. The same exact methods can apply to detecting cancerous cells.

Data science can be applied specifically to manufacturing and many other fields.  A lot of data science goes into process control and manufacturing control. If you scale it to the entire operation and the workflows that go into it, like marketing, then you're still going to have to sell your product. You have to know who your customers are. You have to be efficient with how you distribute supply chain management. That's a massive undertaking if you don't have someone looking at how things are flowing and providing tracking and efficient ways of optimizing your resources.

Basically, you can apply data science anywhere. Good data science always provides actionable insights. If you can measure something, you can probably gain an insight into it, and then if you have an actionable insight, that's what closes the loop.