Introduction to Jakub Czerny, one of our Machine Learning Engineers
Hello Jakub, we would like to know more about you since you joined the Parashift team relatively recently. Could you start by giving us a little wrap-up of who you are?
Hi there, I’m Jakub and originally, I come from Poland, but for the past 8 years I have been living all over Europe – technically Switzerland is my 7th country. Along the way, I worked at a few different places ranging from small- and mid-sized startups to big enterprises, and probably you can guess which ones I am more into. I love working on challenging problems but also being able to see my work go into a product that impacts thousands of people every day.
Very cool! As a person who was working for small startups but also big companies, and in different countries… You must have some experience! Great! What is your role at Parashift?
I work as a Machine Learning Engineer and my daily tasks revolve around improving the models – this includes document classification, information extraction, or things like the latest feature we deployed – an improved document enhancement that attempts to improve the quality of the scan/picture.
Since we are working on a novel problem, my work includes a lot of experimentation but also challenging the ideas and ways of how we are doing things right now.
How did you get into AI and machine learning?
It all started with my bachelor’s degree. Back then, in 2013, the field of machine learning was nowhere near as popular as it is today, but I took a leap of faith and completed a bachelor’s degree in Knowledge Engineering, a combination of applied mathematics, computer science, and fundamentals of machine learning. The funny thing is that I actually graduated with a degree in Data Science, not Knowledge Engineering. In fact, the degree has since been renamed because the term never caught on and people just didn’t know what “Knowledge Engineering” was.
Really funny! Never heard of it before. But now… Let’s put you on the spot; what is the accuracy of our Deep Learning powered AI system?
I think this will trigger most ML practitioners for 3 reasons. In practice we rarely use accuracy as the performance metric – most of the time it is not an appropriate way of measuring how good the model is, though this is what many people from outside of the field expect. Also, the term AI is not really used by the experts. What we work on are machine learning models, which come in many different flavors. We see AI as more of a buzzword. Lastly, not all models that we work with are from the family of deep learning, there are many tasks where this is not an appropriate approach – simply an overkill for the problem.
Here’s a diagram showing the difference.
Thanks for sharing. I guess it makes things much more understandable. At least for me. Now, let’s move on. What are the two challenges in AI that captivate you the most?
I’m very excited about the computer vision approach Tesla is using for its autonomous features. Yes, yes, this is a superhype topic, but what I find particularly interesting is that Tesla doesn’t rely on lidars to estimate depth dimension (distance to objects), but instead uses standard cameras and machine learning to estimate that information. That’s a big difference from competitors and is kind of a showcase for what ML algorithms are capable of. Really cool stuff.
The other is generative models; models that are capable of synthesizing images, videos, or other kinds of data. Training such models is generally difficult and there are many challenges to it, so knowing how it works under the hood makes me appreciate it even more. I think we all should keep an eye on the development of that field as it certainly will affect our lives in the near future.
Speaking of the future; what are your expectations about the future of Parashift?
I absolutely sign with both my hands under the mission and long-term goal that Parashift has set. I really believe we are into something great here. Even though I have not been here for a long time yet, I can already spot many improvements made to the product and see solved pain points of our customers. Now, imagine what will become possible with time once we grow the team. ????
Great to know! With that, let’s come to a final question. What does general intelligence mean to you, and how closely related do you think is the problem of general intelligence and document capture?
General intelligence means being able to perceive a possibly unknown environment and still perform the expected task. Basically, for me, it means being able to understand the environment and interact with it in a meaningful way.
When it comes to document capture, we do not really need something that can sense the surrounding world and interact with it (at least for now), but instead having knowledge about all sorts of documents and files. I see this as one of the skills that a general intelligence machine would possess.
Interesting. Thanks for sharing your perspective. And thank you for taking the time today. We’re all glad to have you on board!