Sid's innovation superpower? Balancing machine learning with data analytical skills in order to unexpectedly delight listeners. Fun fact? You probably didn’t know that Sid was a member of the blackjack team that inspired the book Bringing Down the House and the movie “21”.
Tell us about your current role and responsibilities. Why did you choose to join your current company?
I head the data science org for listeners at Pandora. Our mission is to build, improve and innovate listener experiences across all platforms at Pandora. Pandora wasn’t necessarily in my sights; candidly speaking, I wasn’t looking for a change in my former position. But once I started having discussions with them, it was clear that this was a place I could easily find a sense of purpose.
Music is in our soul, and folks work here simply because they want to be here and do right by listeners as well as artists. This is reflected in our proprietary Music Genome Project (MGP), the largest, richest music content database in the world. Over the last 18 years, Pandora’s in-house music analysis team has hand-labeled millions of tracks in our catalog based on over 400 specific attributes, which serves as a foundation for machine learning to expand the coverage to the rest of the catalog. The MGP, in combination with billions and billions of user interactions, make for a dataset that couldn't get any more intellectually appealing to a data scientist. Plus, we have a killer data science team with backgrounds spanning from computer science, to astrophysics and music information retrieval! It’s truly a wonderful place to be.
What are you working on now that you think is innovative?
What big learning moments have you had in your journey? Did you have any notable mentors?
I got my PhD in theoretical work in the field of Chemical Physics from UC Berkeley; a far cry from what I do today. :) But my learning started very early on. My advisor, Professor Reimer, who I look up to still, drilled in me the notion that how you communicate your work and educate is as important as the work itself. This is one of the pillars of data science - socializing the context, the problem, your methodology, your results. During college, I also happened to be a part of the blackjack team that was eventually featured in the movie “21”. This experience gave me the ultimate thrill of applying math to solve problems in real time. I must admit that there’s no traditional formal path to becoming a data scientist.
Right after grad school, I joined a company called DemandTec as a data scientist. But the term “data science” hadn’t been coined back then, so let’s just say scientist. DemandTec used econometric models on big point-of-sale data for large retailers and helped optimize prices and promotions. I moved on to managing data science teams very early on, and found quickly that I find immense satisfaction in managing my teams. I see it as a service - it’s my responsibility to help everyone grow, solve interesting problems, make meaningful progress, and advance in their careers. DemandTec was eventually acquired by IBM.
Soon after, I joined a startup called Freshplum that had just come out of Y Combinator. At Freshplum, we used machine learning to personalize the shopping experience for visitors in real time; I ran data science and account management. Freshplum was acquired by TellApart, which was essentially doing the same thing but in the ads space. TellApart was a lean and mean machine highly acclaimed in the valley for it’s technical chops and talent. TellApart was eventually acquired by Twitter, which is where I was before I found my new home at Pandora.
I haven’t had any mentors per se, but there are two people in the valley I admire immensely - Shishir Mehrotra, who is building a company called Coda, and Dan Zigmond, who currently runs data science at Instagram. Every time I meet them, I find it to be a learning moment. One notably, that a sense of purpose is the most important ingredient in being happy in your career. Data science is not just about applying math to a carefully curated dataset. Most of your time goes into understanding the context of the problem, and especially wrangling with massive unstructured unclean datasets. So without a clear sense of purpose, the day to day frustrations seem that much more insurmountable, and lead to discontent.
How do you pick and develop the talent on your team? How do you ensure there is collaboration?
I know there’s a perpetual debate on this subject, but I believe in building a centralized data science team rather than data scientists being embedded in different product and engineering teams. This gives you the ability to build a diverse team that is essential to its success, and also hire at different skill levels facilitated by establishing a strong mentorship program. It also leads to better collaborations as scientists in the same team can be hired and trained in a way that are complementary to each other. I cannot overemphasize the importance of building a diverse team. At Pandora, for example, our audience is extremely diverse with unique musical tastes. It’s important that our data science team reflects that diversity in order to build delightful products for them.
“I cannot overemphasize the importance of building a diverse team. At Pandora, for example, our audience is extremely diverse with unique musical tastes. It’s important that our data science team reflects that diversity in order to build delightful products for them.””
In terms of who we look for, Pandora is clearly a data-driven product. To build something of that nature, you can only go so far by sitting in a room and endlessly discussing ideas in abstract. The right way to do this is by building prototypes that one can see, test, and get a feel for. This approach leads to much more focused discussions, and has resulted in some of Pandora’s most successful features like Thumbprint Radio and Personalized Soundtracks. To enable this, we require our data scientists to be builders as well as theoreticians; qualities we definitely look for.
We ask a lot from our data scientists, but as Reid Hoffman alluded to, there has to be a form of mutual investment. As an example of career development, we invest in our data scientists by sponsoring attendance to two or more conferences a year for every scientist. Pandora has always had a strong presence at the premier recommendation systems conference, RecSys, and the music information retrieval conference, ISMIR, via presentations and workshops. The goal is twofold - a) education - learn from others and stay up to date with cutting edge technologies, b) ownership - showcase our own learnings to enable advancements in our fields. At the end of the day, for scientists, these two aspects of research are very important because they serve as motivation and eventually lead to better products. All these steps and care have also led to the listener data science team having low to almost no attrition.
Where do you see yourself in five years?
My honest answer to this has always been - that’s too far for me to foresee. But for the foreseeable future, I want to be a part of the Pandora journey. Pandora is a machine learning company with music in our soul. We have a very clear sense of purpose: to delight our listeners - wherever they are, who they are with, what they are doing and what mood they are in - with a diverse set of audio experiences. And we certainly have the data and the machine learning capabilities to do just that better than anyone else. I expect my teams to not just inform the business, but to drive the business to get there.
Favorite place to vacation? I am a parent now. We go where our kid wants to go. :) Currently it’s anywhere warm with a sandy beach. Works just fine!
If you were a superhero, what would your special skill be? Well, our 7-year-old, who has just figured out the concept of consequences for his actions, recently has been raving about a dream he had about a superhero called, “The Consequence Man.” Once I understand what that is, that might be my calling.
Name something that most people don't know about you. Sometimes I listen to A-Ha. When nobody is watching.
Watch Sid talk about Machine Learning and How to Build Your Data Science Team.