From Developer to Successful Machine Learning Entrepreneur: Aparna Dhinakaran, Co-Founder, CPO of Arize (Part 1)

We have a huge audience of developers, engineers, and programmers who want to transition to becoming successful entrepreneurs.

This conversation explores the journey of such a developer. Fantastic story!

Sramana Mitra: Let’s start at the very beginning of your journey. Where are you from? Where did you grow up? What kind of circumstances?

Aparna Dhinakaran: I was born in Chennai, India and my parents immigrated here when I was six months old. They immigrated all over – Boston, Chicago, Illinois. They settled down in the Bay Area. I really liked Math. That led me to go do a computer science degree at Berkeley.

Sramana Mitra: What year did you graduate?

Aparna Dhinakaran: I graduated in 2016.

Sramana Mitra: What did you do when you came out?

Aparna Dhinakaran: My first job was with Uber. In retrospect, I’m so glad that I did that. Uber was going through its growth phase. The company was doubling every six months. New orgs were being created. There’s a reorg every six months. The cool thing for me was, six years ago, all of the things that we see now on Uber like surge pricing, routing, or forecasting were just starting to become machine learning models.

There’s even a guy in my team in the beginning who was in surge pricing. A lot of those were just these teams starting to build infrastructure to put them into production. As I think about this, there’s probably only a handful of companies out there that was using machine learning at that time in their core products. Companies like Uber were doing it in real-time.

Sramana Mitra: You were part of the machine learning team there?

Aparna Dhinakaran: Yes. Eventually, some of the stuff that we built in the ML teams became a part of the broader Uber ML platform that everyone now sees. They’ve open-sourced a lot of it. You have a lot of ML companies now who are trying to reassemble their stack off of what Uber had built. They were trying to educate how to get an ML platform set up at an organization.

Sramana Mitra: How long did you stay at Uber?

Aparna Dhinakaran: Close to three years. I spent some of it working on the modeling side. The majority of it was on putting those models into production. How do you handle the complexities of making ML more like software?

The way that we build software today is very engineering-focused. The way that we build machine learning is still research-focused. It feels more like you’re in a research environment with data scientists playing around with those lab-type notebooks. That’s where I spent a number of years at Uber. Uber is probably one of the first places where I started thinking about my idea.

I was the persona that my startup now is for. I was an ML Engineer. I put models into production. I felt the pain when models weren’t doing what they were supposed to be doing. If somebody ever complained that incentives were off or forecasts were off, I literally had to go back and answer to them. I didn’t have any of the tools to be able to go do that.

You think about all regular software. They have all sorts of tools. I just got off a call with someone from Grafana. There are tools from Datadog and New Relic that help you troubleshoot and answer infra problems, but there are no solutions to help you answer when your models break. Feeling a problem yourself is invaluable.

Sramana Mitra: Yes, and you really know the nuances of the problem at a very deep level.

Aparna Dhinakaran: I spent a lot of time building what I am trying to build today.

Sramana Mitra: What path did you take into entrepreneurship?

Aparna Dhinakaran: There are probably two big things. One was how do ML teams work that are different from just regular software teams. There are these nuances in creating a new category. It actually takes figuring out, “Why can’t I just take the tools that worked before?”

The natural step for me is to think why don’t the tools for software work for me in machine learning. Prove to yourself that this is a new category. Why is this different? How is this different? Why don’t the tools that are used in other categories work for you here?

The second big piece is how to troubleshoot. A model is not doing what it’s supposed to be doing. What’s causing that? How do I, as an engineer, go through the workflow and fix that? How do I fix the problem myself and then figure out how to put it into production? If you don’t do it yourself, it’s very hard to build a product around that.

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