A Q&A with Vladimir Rybakov Head of Data Science at WaveAccess

Jakub Czakon, Senior Data Scientist at Neptune, talked to Vladimir Rybakov, Head of Data Science at WaveAccess, to discover how to solve problems fast under extreme pressure, how taking responsibility can help you progress in your career, why asking for help may be one of the most important skills to obtain, what were the key milestones of Vladimir’s career path, and much more. Get the interview's key takeaways in our blog post, while the original podcast can be found on Neptune blog.

J.C.: Hi Vladimir. What Data Science projects are you working on?

V.R.: Thank you for asking. It's difficult to actually define any specific field that we focus on because we are currently involved in all sorts of projects within text mining, computer vision, forecasting and data analysis in general. But the most dominant direction right now is applying Machine Learning (ML) and Data Science to CRM systems. We receive many requests about using ML for consumer data from our clients: these could be huge datasets that companies need to process manually, which is the area prone to human mistakes, where automation can help most effectively.

Another significant area is Text Mining that is closely related to the CRM systems. We are also involved in Computer Vision Projects — practically, the largest ones. 

We are working on an exciting new project where we have to search for illegally deforested areas and fields that are not being used properly, based on satellite images analysis. Certainly, similar systems are already in place, but the project involved lots of specific requirements, so we had to develop the system from scratch. Actually, it is a common situation: in my experience, 60-70% of open-source solutions are not very useful when it comes to real-world problems. The results that they showcase (if they do it at all) are usually cherry-picked to make a good impression, but nothing more.

J.C.: What was your career path?

V.R.: I’ve been a Head of Data Science for the last 2 years, and it took me around 5 years to get here. Quite a few people have asked me: “How did you manage to become the Head of Data Science so fast?”. I would like to come back to that in a little bit.

When it comes to my background, I am a Mathematician. I started studying Physics, but decided to change my major and delve into mathematics more deeply. In my fourth year I first discovered the new world of Artificial Intelligence and got really interested in it. I started working closely with my professor, we received a couple of grants and I was fortunate enough to spend my master’s degree in the field of Data Science. 

Later, I was working as a software developer for a year, but didn’t enjoy it very much. We were working on older robust systems, and I didn’t see myself growing there at all. So I made up my mind to try my luck as a Data Science freelancer. As I was working alone, it was quite hard not to have anyone around to get advice from, and at that time there were not a lot of resources I could use to learn. I was engaged in a curious project where I had to classify the person’s personality based on the face and ear shapes. Long story short, we didn’t click with the client so well, so I decided to look for another job.

I was already 25 years old and still quite a junior developer. I managed to get two offers: one of them was from a multinational home appliance manufacturer, and the other — from WaveAccess. I realized that it was never my dream to work for a huge corporation with a lot of bureaucracy, where I would simply become another minor contributor. At that moment, WaveAccess had a little over 100 people on board which I believed would be a way better fit. So, I got the job as an Algorithmist where I had to develop algorithms in Java that would process graph data.

After a while, I approached my supervisor Alexander to ask him: “I know Data Science pretty well, are there any projects that I could help you with?”. His answer was “Not really”. Some time later, Alexander came back and said: “There is a small project for a large company where we need to create a Proof of Concept. This is your chance. Can you deliver it in a week?”. I worked hard and completed it on time. And that’s how my Data Science career started. I was the first Data Scientist at WaveAccess, working hard on many projects and learning every step of the way.

After some time I was able to hire the first team member to implement a huge project where we needed to predict the routes of moving objects. It was both exciting and exhausting. It was a coding marathon: once I've stayed at the office for 40 hours straight without sleep, but was able to deliver the customer demo on time. We continued to expand the team, and the Data Science department now has 20+ experts.

Vladimir Rybakov, Head of Data Science, WaveAccess

J.C.: What surprised you the most about your current role?​

V.R.: I did not realize how much additional administrative duties I would need to perform: planning, budgeting, managing the team, presenting the results. I quite often find myself at the meetings analyzing the team’s results, trying to figure out what is needed, to be improved upon, in order for my team to succeed. I spend time giving presentations to clients, or interviewing new candidates. Much of my time is also spent working with our sales and marketing team, as it is necessary for them to be well-versed in projects we have implemented or are in the process of implementing. 

I would say that if you are not a person that likes to be the center of things, then Head of Data Science may not be a proper role for you. You have to feel comfortable with communicating and holding negotiations everyday.

J.C.: What are the crucial skills to be a good Data Science leader?

V.R.: Communication is definitely one of them, but, in my opinion, the most important skill a leader should have is the ability to always be ready to implement a backup plan. If something goes wrong, you are the one  that must come up with the best solution.  Moreover, some of the problems that you will face will be brand new and therefore hard to completely understand. When such a situation occurs I first try to break the problem down to some basic concepts and drop all the specifics. It helps get to the core and propose a way out based on your experience. Then I pose lots of questions to a client to understand the problem from their perspective. Finally, I try to suggest a general solution without going into details. When a client has agreed to proceed, it's at this point, I start thoroughly digging into details. From my experience, there is no way to grasp the project completely before you actually start working on it. 

Getting back to communication, it is absolutely crucial that both a client and tech consultant are speaking the same language. Clients want to be sure that you are taking responsibility and full control of the process. By the way, this is what we are involved in right now. WaveAccess is running special workshops with a series of sessions. There is a large number of companies out there that want to introduce Data Science to their processes, but they lack the understanding of what exactly can be done and what is required to build models. This is where our consulting sessions come into play.

Another crucial thing is your willingness to take responsibility. Don't hesitate to understand how your solution brings more value to the business. Ask your executives: “How can I take more responsibility on the projects I am involved in? What do I need to do to get  a promotion?”. There is no guarantee you will immediately get the desired answer, but you can put yourself in a position where positive things can happen to you. Theу say that “only those who work hard get lucky”. Certainly, a fair and sensible boss plays a crucial role as well. I was lucky to have one like that.

One more thing that I feel is really important is staying up-to-date on the latest ideas, innovations and research. You must be on top of all that.

J.C.: What was the most difficult skill for you to learn?

V.R.: These were issues related to scaling the team of 2-3 people to a bigger unit. At first, I struggled against introducing bureaucracy and administrative matters to our department just because it’s against my nature. Although, at some point, I understood that it was necessary. 

Another thing that I discovered was that, as you grow in your organization, you manage more and develop less. It took some time to get used to the thought that I was no longer a developer, but a manager. It became crystal clear when I began taking post-graduate courses in management where I had to get a better understanding of how  business works and to advance my project management skills.

I would suggest  management courses to anyone hoping to succeed in their Data Science careers. In 5 or 10 years those Data Scientists who devote more time to understanding the business value of what they are creating will stay relevant regardless of what happens with AutoML and other things like that. You need to contribute more than just fitting models.

J.C.: How do you keep yourself informed?

V.R.: I follow a bunch of smart people on Twitter such as Andrej Karpathy (@karpathy), Jeff Dean (@JeffDean), Dustin Tran (@dustinvtran), Andrew Ng (@AndrewYNg), Nando de Freitas (@NandoDF), Pieter Abbeel (@pabbeel), Andrew Trask (@iamtrask), Ilya Sutskever (@ilyasut) and Trent McConaghy (@trentmc0). Whenever a new exciting paper comes out — I will  find out about it from their tweets. I also read Medium a lot. To me, it is probably the best online resource that provides high-quality, relevant articles not just looking at them from an academic point of view. There are also a few YouTube channels that I adore such as Two Minute Papers. Recently, they have posted a video about the Open AI paper where the concept of model surgery was observed. The authors introduced an approach where one can look inside the model and change its architecture without stopping the training process. I found it fascinating indeed.

J.C.: What should junior specialists primarily learn to be successful at Data Science?

V.R.: While interviewing potential candidates I see a lot of them taking many online courses and believing this will make them good Data Scientists. I do think that practice is the key to a successful career as a data scientist. Try carrying out your own project, or go to Kaggle and try implementing things that you have learned. It is a great resource that provides an option to not only participate in projects under close-to-real conditions, but also to learn through discussions and kernels of other people that are published for all competitions.

There is one more principal element, it is not  technical but an equally valuable one. I firmly believe that you should be a good person. Our Data Science department’s structure is quite horizontal, and although the final decision still rests with me, the input from others is crucial.  You should set an example for your team both in terms of knowledge and communication, establish good standards and values, and create a positive workplace culture.

J.C.: Why is it so important to learn how to ask for help?

V.R.: We have a “30-minute rule” in our department. If one of us faces a problem and can’t find a solution to it in half an hour's time, then he or she should ask others for help. It’s not that the task should be completely managed in 30 minutes, but this slot is enough to look for materials online and consider the possible options. After that period your efforts can easily start to move in circles and waste time.

I want my team to feel comfortable asking for help. I personally often request them to explain papers and ideas to me, as there is no way I can stay informed about all the latest tricks.

There must be a clear balance here. Asking for help all the time won’t facilitate you growing as an expert, while trying to solve everything on your own will lead to wasting too much precious time unnecessarily.

J.C.: Do you have some final thoughts to sum up?

V.R.: I came across a meme recently saying “Data Science is pain. If somebody tells you differently — they are selling you stuff”. To sum up, I’d note the most important element which is to have fun and enjoy your work — regardless of whether you are a Junior Specialist or a Head of Data Science. I strongly oppose those who start doing something just because it is a lucrative domain. Do what you love and the success will come naturally.