"Fundamentals Are All There Is": An Interview with Senthil Gandhi, Award-Winning Data Scientist at Autodesk
By Metis • December 14, 2017
We had the pleasure of interviewing Senthil Gandhi, Data Scientist at Autodesk, a leader in 3D design, engineering, and entertainment software. At Autodesk, Gandhi built Design Graph (screenshot above), an automated search and completion tool for 3D Design that leverages machine learning. For this pioneering work, he won the Autodesk Techie Innovator of the Year Award in 2016. He took some time to chat with us about his work and about the field of data science in general, including advice for aspiring data scientists (hint: he's big on the fundamentals!).
Metis: What are the important skillsets for a data scientist?
Senthil Gandhi: I believe fundamentals are all there is. And when it comes to fundamentals it is hard to have more mathematics under your belt than you need. So that is where I'd focus my time if I were starting out. Mathematics gives you a lot of great tools to think with, tools that have been perfected over millennia. A side effect of learning mathematics is learning to think clearly – a side effect that will be directly applicable to the next most important skill on the list, which is to be able to communicate clearly and effectively.
Metis: Is it important to specialize in a specific area of data science to be successful?
Senthil Gandhi: Thinking in terms of "areas" is not the most effective mindset. I believe the opposite. It is nice to change your area from time to time. Elon Musk doesn't think rockets were not his "field." When you change areas, you get to carry great ideas from your old area and apply it to the new domain. That creates a lot of fun accidents and new possibilities. One of the most rewarding and creative spells I had in recent times was when I applied ideas from Natural Language Processing, from when I worked for a news company, to the field of Computational Geometry for the Design Graph project involving CAD data.
Metis: How do you keep track of all the new developments in the field?
Senthil Gandhi: Again, fundamentals are all there is. News is overrated. It seems like there are 100 deep learning papers published every day. Certainly, the field is very active. But if you knew enough math, as in Calculus and Linear Algebra, you can take a look at back-propagation and understand what is going on. And if you understand back-propagation, you can skim a recent paper and understand the one or two slight changes they did to either apply the network to a new use case or to increase the performance by some percentage.
I don't mean to say that you should stop learning after grasping the fundamentals. Rather, view everything as either a core concept or an application. To continue learning, I'd pick the top 5 fundamental papers of the year and spend time deconstructing and understanding every single line rather than skimming all the 100 papers that came out recently.
Metis: You mentioned your Design Graph project. Working with 3D geometries has many difficulties, one of which is viewing the data. Did you leverage Autodesk 3D to visualize? Did having that tool at your disposal make you more effective?
Senthil Gandhi: Yes, Autodesk has a lot of 3D visualization capabilities, to say the least. This certainly turned out to be handy. But more importantly during my investigations, a lot of tools had to be built from scratch.
Metis: What are the big challenges in working on a multi-year project?
Senthil Gandhi: Building things that scale and actually work in production is a multi-year project in most cases. Once the novelty has worn off, there is still a lot of work left to get something to production quality. Persisting during those years is key. Starting things and staying with them to see them through involve different mindsets. It helps to pay attention to this and grow into these mindsets as it becomes necessary.
Metis: How was the collaboration process with the others on the team?
Senthil Gandhi: Communication between team members is key. As a team, we had lunch together at least twice a week. Note that this wasn't required by any top-down communication. Rather it just happened, and it turned out to be one of the best things that accidentally helped in pushing the project forward. It helps a lot if you like spending time with your team members. You can invert this into a heuristic for finding good teams. Would you like to hang out with them when it is strictly not necessary?
Metis: Should a data scientist be a software engineer too? What skills are important for that?
Senthil Gandhi: It helps to be good at programming. It helps a lot! Just like it helps to be good at math. The more you have of these fundamental skills, the better your prospects. When you are doing cutting-edge work, a lot of times you'd find that the tools you need aren't available. During those times, what else can you do, than to roll up your sleeves and start building?
I understand that this is a sore point among many aspiring data scientists. Some of the best Data Scientists I know aren't the best Software Engineers and vice versa. So why send people on this seemingly impossible journey.
Metis: What skills will be important in 10 years?
Senthil Gandhi: If you have been carefully reading so far, my answer to this should be pretty clear by now! Predicting what skills will be important in 10 years is identical to predicting what the stock market will look like in 10 years. Instead of focusing on this question, if we just focus on the fundamentals and have a fluid mindset, we could move into any emerging specialties as they become relevant.
Metis: What's your advice for data scientists that want to get into 3D printing technologies?
Senthil Gandhi: Find a problem, find an angle in which you can approach it, scope it out, and then go do it. The best way to get into anything is to work on a relevant specific problem on a small scale and grow from there.
Going through a data science bootcamp is an intense experience for everyone involved. Students work at a breakneck pace unparalleled in other learning environments, absorbing new and difficult concepts and skills, and applying them to projects starting as early as week one. All the while, instructors shoulder the hopes and fears of their cohorts as they guide and teach them over the course of 12 weeks. For interested individuals, teaching a bootcamp can have tangible career benefits related to industry goals and aspirations. Read how here.
As we approach another Final Application Deadline (Monday, March 5th!) for Data Science Bootcamps in Seattle, San Francisco, Chicago, and New York City, we sat down with Metis Sr. Data Scientist Jonathan Balaban to ask him about the value of the 12-week program. In this Q&A, he talks about his professional background, why he wishes data science bootcamps had been around when he was learning and pursuing a career, and what students will get out of their time at Metis.
Paul Trowbridge, instructor of our upcoming Live Online Statistical Foundations for Data Science & Machine Learning course, discussed the need for a firm stats foundation, talked about his career, and more during a recent Q&A.