From Developer to Successful Machine Learning Entrepreneur: Aparna Dhinakaran, Co-Founder, CPO of Arize (Part 2)

Sramana Mitra: So did you quit Uber?

Aparna Dhinakaran: The reason I left Uber was to go to a PhD program. No one in my family had ever gone through higher education. When I had applied to Uber, I had an offer from a PhD program in Computer Vision. I had a ton of debt, so I had to first go to work for a couple of years. I’m really glad I did because nothing beats real-world experience. I left Uber to do a PhD at Cornell.

I should tell you a quick side story because it’s related to what made me decide to leave Uber. While I was at Uber, I was on this reality TV show called Amazing Race. They pick about 10 teams. You get to go compete and race around the world for a million dollars. That’s the premise of the show.

Sramana Mitra: Race around the world doing what?

Aparna Dhinakaran: There are these clues in all these countries. Think about it as a big scavenger hunt. Every leg of the race, you’re going for a different country. Every leg, one team is eliminated. The final team wins the million dollars.

I did that while I was at Uber. It was a life-changing experience. That made me take more risks in my day-to-day life. Every single day that we were on that show, you were waking up and doing something completely different. I came back from that experience and thought that I wanted more of that in my day to day. I took a bet on myself and decided to get my PhD. That’s what got me to quit my job.

I went to Cornell. While I was there, the biggest thing that it did for me was it put me in a new environment and threw a bunch of problems at me. The first problem that I latched on to was this big question around AI fairness. It’s very much talked about now. How do you detect fairness? How do you check for these things?

Because it was a research place, they were thinking about everything from a more academic perspective. They were working on how to measure fairness. I came in from putting models into production. In the real world, people are just trying to get this shit to work. You don’t even have the infrastructure to get this stuff to work. I didn’t have the tools.

There was this moment of dissonance. Academia was so far ahead in evaluating fairness. There was this gap of if we really want that stuff that we’re talking about in research to happen in the real world, there’s no way we’re getting there. I didn’t see a path.

Going back to that point of did I quit my job to be an entrepreneur, I don’t think an entrepreneur’s path is linear. I feel like you have different points in your journey where you’re getting to the same idea. You need those different points.

Sramana Mitra: You got two ideas. One at Uber where you decided that the machine learning infrastructure didn’t exist. At Cornell, you decided that the ethics infrastructure didn’t exist to be able to implement this in production.

Aparna Dhinakaran: Exactly. It took a while to probably realize they’re both the same. They both meet at a point. On the ML infrastructure side, there are no tools to help you do observability and even just stop being blind. I can’t emphasize enough. From the moment you wake up to the moment you eat lunch, there are a thousand models that are touching you – ride-sharing apps, weather apps, food delivery. When you wake up in the morning, you open your news app. The news app has recommendation models.

Sramana Mitra: At Cornell, your idea was crystallizing further. Did you quit your PhD program?

Aparna Dhinakaran: Yes. I spent a lot of time in shipping all of these models blind. That was my thesis. We don’t have any way to troubleshoot. Over at Cornell, the thesis became we’re shipping them blind and that can have detrimental consequences.

I think there was an urgency that came about. Think about a company that decides who to give a loan to. If my model is declining certain individuals, I wouldn’t be able to know that. I wouldn’t be able to figure that out. All of that stems from a lack of observability.

At that point, the Y Combinator application was due. It popped up on my LinkedIn. I had really nothing. I had a few mocks of what the product would look like but not much. It was due at midnight that night, so I submitted an application. I got called back for an interview. I did the interview. In that one week, I put together a mock of what the product would look like. I pitched it to YC and got told that day that we were accepted.

This segment is part 2 in the series : From Developer to Successful Machine Learning Entrepreneur: Aparna Dhinakaran, Co-Founder, CPO of Arize
1 2 3 4 5 6 7