interviews

Adam Gibson, Skymind

Elyssa Macfarlane
Adam Gibson, Skymind

When Adam agreed to do an interview with MoT, I was psyched. He’s currently located in Tokyo and directing a large and diverse team of deep learning engineers- so I appreciated the time he was dedicating to doing an interview. Then I found out that he is the same age as me, and I was baffled.

As the co-founder and CTO of Skymind, Adam has some interesting experience under his belt. At the age of 21 he dropped out of Michigan Tech to found his first start up. In 2014 he and Chris Nicholson cofounded Skymind- a deep learning service for enterprise that employs deeplearning4j, a framework created by Adam. The business entered Y combinator’s Winter 2016 batch and has since acquired a variety of contracts from MNCs that you’ve likely heard of.

When I’m on the phone with Adam, I can already tell that he is an effective person to be around. Everything he discusses is filled with purpose, where he never wastes time by overelaborating on the concepts and opinions that we talk about. He is to the point, while also being critical and honest.

In the instance that you aren’t technical and don’t quite understand what a framework is or how Skymind operates, I recommend checking out this article here.

Interview

Tell me a bit about how you found yourself working in tech. When and how did you initially start programming?

Most programming founders you hear about started when they were 11. I was just a competitive gamer. I started coding in college- I also failed my first computer science class. I learned to like coding my freshman summer and from there devoted time I spent gaming towards learning computer science. It was the best decision I made.

How did you come up with the idea for deeplearning4j?

I started doing machine learning in 2011 and tried a number of startups in machine learning before moving to san francisco in 2013. I had heard about deep learning back then from google and other people presenting on some of the work being done. I did a survey of the market and found that there were no deep learning tools in java. I had a decent background in ML at this point and figured I would build the tool I wanted for myself.

Do you find that the market is saturated with too many open source frameworks? What gives deeplearning4j a competitive advantage?

All deep learning frameworks have the same problem. They focus on the math and serve research and hobbyists. No one thinks about the boring problems in the space like ETL (Extract, Transform, and Load) or integrations with different kinds of databases. The other extreme here are wrappers for deep learning frameworks that hide too many details and break when you try to do something customized. Deeplearning4j is in the middle. Everything is decoupled but there are best practices for integrating the parts to build a real product. This has an advantage, especially in software engineering where you need to test components separately, or maybe swap them out. Deeplearning4j’s main advantage is that it does a ton more than math. You get math, reinforcement learning, ETL libraries and integrations, distributed systems integrations like spark, and models you can easily deploy in a java application with zero hassle.

How is Skymind able to attract and retain their talent? I imagine that experienced deep learning engineers are tough to come by.

They aren’t hard if you hire globally and let people stay where they are. We are distributed across six time zones. On average my colleagues are older… This is a good thing. We retain people by not having the same bro culture many silicon valley startups do. A lot of startups insist on ping pong tables and beers on tap while “living” at work. We focus on real benefits like work/life balance, being able to work wherever you want (offices are optional) and keeping people focused on concrete goals rather than an endless death march.

Can you describe your experience with Skymind at Y-combinator?

YC was great for us. It was more of a “stamp of approval “ though. We didn’t listen to half of the advice from the partners (much of it is conflicting, which they tell you). We didn’t use things like the demo day PR from TechCrunch. Instead we had a customer interview published in wall street CIO where our actual customer base reads.

Are there instances where you would advise a startup to avoid going through an accelerator?

Yes. Understand what benefits the accelerator gives you. Don’t just blindly listen to what people tell you. Even while you’re in the accelerator focus on customers and users (this is what YC does which is great).