Employee Spotlight: Vladimir Iglovikov

Vladimir Iglovikov is a mountaineering, rock climbing, Burning Man-loving scientist who holds both a Master’s and Ph.D. in physics. He has published a number of papers on everything from theoretical physics to the application of deep learning on forensics, medical, and satellite imagery. As he approached graduation, he decided he wanted to explore something new and began looking for industry jobs. To prepare, he began competing in machine learning competitions where he eventually became a Kaggle Grandmaster.

He landed his first industry jobs working as a data scientist at startups, then joined Lyft’ science team. That’s when Level 5 caught his attention. At the time, the self-driving industry was new to Vladimir, but the impact of the problem Level 5 was solving excited him. So, he began to research, study, and continued competing in machine learning competitions. This boosted his skills enough to join Level 5’s mapping team.

We interviewed him to hear more about his journey from publishing papers to developing self-driving cars.

Can you tell us about your role at Level 5?

I work on the Level 5 mapping team. High-definition maps are an important part of the self-driving technological stack. If something can be computed ahead of time, cached, verified, and copied into the car, it will perform more effectively. Things like traffic lights don’t appear and disappear, and traffic lanes don’t emerge from nowhere. These are things that we can map ahead of time.

In order for the maps we create to be useful and scalable, they must be accurate, high-definition, and up-to-date across different geographies. The way we do this is to leverage machine learning techniques, applying them to the data from lidar, radar, camera, and other sensors. I’m working on delivering those models.

How long have you been at Level 5?

I’ve been with Lyft a total of 1.5 years. I originally joined the research science team on the rideshare side of Lyft to make maps more detailed using computer vision techniques. Six months later, after talking with the Level 5 self-driving team, it was apparent to me that this was the place I wanted to be. The project goals, product fit — everything clicked. In April of last year, I moved to Level 5.

Tell us a bit about your background before Lyft.

I got my Master’s in Russia in theoretical high-energy physics, which is a branch of physics that studies the nature of particles that constitute matter and radiation. After that, I moved to the US because I believed the strongest graduate schools were here. I then studied theoretical condensed matter physics at UC Davis. It was a lot of math, coding, statistics, data analysis, and mostly understanding how materials are made at the elementary level.

When graduation came, I had to decide what I wanted to do next. My options were to go for a postdoctoral position or to get a traditional software engineering job. I did not want to stay in academia, and after coming from studying such fascinating topics in school, frontend/backend positions in the industry did not sound anywhere near as exciting to me.

My friend, also a physicist, told me about data science, which had a significant overlap with my background in physics. I looked into it, started studying, and eventually went to work for a couple of startups as a data scientist before transitioning to Lyft.

The main challenge that I faced when applying to data science positions was not the lack of technical skills, but the absence of the standard keywords on my resume, like names of the big companies or publications at relevant conferences.

What helped with my journey (and my resume) was developing my machine learning muscles through online data science challenges. I got into them four years ago through Kaggle. Somehow, maybe it was my physics background or my competitiveness, I got top places in a number of them. I participated in more than 70 at Kaggle, Topcoder, MICCAI and CVPR; won cash prizes; and eventually got the title of Kaggle Grandmaster. The combination of my skills from academia, industry experience before Lyft, and the knowledge gained from competitions helped me to get a position at Level 5.

After spending some time in the industry, what advice would you give to your past self to prepare for this?

Job searching is hard. It is not enough to be smart, good at math, and able to publish and present papers. You need a much more diverse set of skills. Some can only be developed by working on real products in the industry, but plenty of them can be at least partially learned while you are still in academia.

For example, mastering Leetcode for algorithms and data structures, and Kaggle for machine learning before graduation would have simplified my job search a lot. These platforms provide an excellent learning environment, and it would have been wise to start investing time into them much earlier.

To solve for lacking coding skills, I would recommend contributing to open source projects. Working on an open-source project is still not the same as writing production-quality code on an everyday basis as a part of the large team at a company with high coding standards. Nothing can replace this experience. But working on open-source projects can prepare you relatively well.

Technical skills are essential, and working on small projects or research teams with like-minded people who share those similar skills is great — communication is easy and you all speak the same language — but it’s wasn’t enough to prepare me for the industry. When the problem becomes large, the total number of required skills becomes so big that you end up on a large team made up of people with entirely different backgrounds. That is where your soft skills become essential.

The challenge is that to develop the soft skills that are needed in the industry, you need to be in the industry. Internships are a great way to solve for this. Here at Lyft, we get plenty of excellent software engineers who first join us as interns and later as full-time employees. As a member of the physics department, I naively didn’t even know that I had such an option. Knowing what I know now, I believe that it would have been highly beneficial to spend every summer of my graduate school in different internships. This would have boosted my coding skills, soft skills, networking skills, and even my physics research.

Can you tell us more about your experience transitioning from academia to working in the industry?

Getting my first offer was hard. After the first six months, I started questioning if I would ever be able to do it. When I finally got a position, my first few months at work were a shock. I was used to collaborating with the like-minded people that work on highly theoretical fundamental questions of nature. The dynamic environment of Silicon Valley, where tasks are much more applied, planning is much shorter-term, and the team is so diverse was foreign to me. It was painful and exciting at the same time. I was used to being an expert in the field, and this typically meant that I didn’t learn that much on a daily basis. The industry was different. I was learning so much and met so many bright people that all the stress that accompanied the transition was worth it.

My time in graduate school was great. Fundamental science is my true passion. But I am happy that I moved to the industry, especially now that I’m working on autonomous vehicles.

Would you recommend the self-driving industry to someone coming from an academic background like you did?

Absolutely. The autonomous industry is very challenging from both the engineering and scientific perspective.

When I look at the scientific pain points in the self-driving industry and compare them to the academic research in the autonomous space, I see a mismatch between what is needed and what the research community is doing. This could be because of constraints on computational resources, lack of access to massive labeled datasets, and lack of first-hand experience with what works and what does not in the self-driving industry.

When you join the self-driving industry, you will work on projects that are interesting, impactful, and that will push you pretty hard out of your comfort zone in terms of how much you think you know about science and technology.

In the self-driving industry, you’re in a constant state of learning — feedback from the universe comes so quickly that you’re regularly forced to learn new technologies. Being in such a fast-paced environment has taught me more than I ever would have learned in academia in the same period of time, which is great because I’m a very curious person. The informational hunger that I get if I do not process new information can drive me crazy, but since I joined Level 5 I get more than enough of it. Here, there’s a lot more to do on any given day, I’ll give you that. But from a learning perspective, it’s worth it.

What keeps you excited to come to work?

I could probably find a lot of good reasons, but all of them can be summarized in a simple statement: I enjoy it.

I enjoy working here because I find the challenge drastically more exciting than any other industry, and the people are great to work with. The problems we’re working on and the challenges we’re facing are changing humanity. At the same time, the problem is complex, and the only way to progress here is to hire the best talent. They may not all be experts in self-driving, but if the person is smart and has a strong background, they can do this. That’s what we look for at Level 5. We are picky about who we hire, and we have a good team; some who have been doing this for many years, and some like me who transitioned from a different background. People like this do well here through passion and hard work.

What has been something unexpected or challenging that came up during your time at Level 5?

At first, everything was challenging. When I joined Level 5, I was good at deep learning and computer vision, but that’s a relatively small part of the stack. For the first few months, during the day I was doing what was required at work, and I spent evenings and weekends learning more about self-driving technology. At the same time, I do not remember any single lunch at work in the first three months that I spent alone. To get up to speed, I tried to talk to as many people from different teams as possible. It was worth it. Now, a year later, I believe I have a pretty good understanding of the self-driving technology, what is possible, what the limitations are. Being able to understand the big picture helps a lot with everyday tasks.

Has there been a particular “win” in the work you’ve done here?

We needed to prove to leadership, the industry, and ourselves that we could deliver something concrete. For us, it was the employee pilot that we launched in December of 2018. It entailed hard deadlines, intense coordination between teams, and a lot of hard work. I watched the mapping team evolve into a highly optimized function. The pilot was a win for all of us and proved that we are capable of building self-driving transportation.