Traditionally, IT has been a male-dominated field. In major IT companies, there are at least four men for every female employee, and only 26.5% of leadership positions are held by women. To support women in tech, the IT community regularly organizes dedicated events. One such event was the Girls Talk Corporation meetup in Berlin. One of its participants, Mariia Bulycheva, an Applied Scientist at the shopping platform Zalando and a mother of four, shared her experience: how she entered the IT field, what she has learned, and how she supports other women who want to pursue careers in technology.
Mariia, tell us a little about yourself: how did you get into Machine Learning, and why did you choose this field?
I have always been interested in trends and new technologies. I studied at the Faculty of Mechanics and Mathematics at Moscow State University, and towards the end of my studies, finance and consulting became highly popular. Many of my classmates chose finance over IT, and I decided to follow the same path.
After graduating from the Mechanics and Mathematics faculty, I pursued an economics degree to gain a better understanding of finance. Shortly after, I received a job offer from J.P. Morgan Securities. Initially, I worked in Moscow, and later, I was relocated to London.
My career was progressing successfully, I was an experienced specialist with a solid salary. However, the working conditions were quite demanding—I had to be in the office by 8 a.m. and often didn’t leave until after midnight. Eventually, I went on maternity leave and had two children within a few years. This break gave me time to reflect on my future career path. Returning to my previous job didn’t seem ideal, as it would leave little time for my family.
At that time, in 2015, machine learning was becoming a hot topic. Neural networks had been invented long ago, but for a long time, there wasn’t enough computing power to make them truly effective. Then, a technological breakthrough happened, enabling more advanced predictive models. I realized this was a promising direction, especially since I already had a strong mathematical background.
What drew me to this field was the rapid pace of development and the constant need to learn. I love acquiring new skills, exploring tools, and mastering programming languages. Additionally, career paths in machine learning are not restricted—you can choose a managerial track, dive deeper into a specific area of expertise, or even switch between different domains of machine learning. In contrast, finance primarily offers growth through management roles.
To deepen my expertise, I completed machine learning courses over nine months. My university education, combined with knowledge of the latest technologies, allowed me to quickly immerse myself in the field. Of course, making such a major career change was challenging, which is why I now actively support women in IT.
You have a strong academic background in mathematics and economics, which clearly helps in your work. Do you think it is possible to become a strong specialist in the field of Machine Learning without such fundamental knowledge?
I firmly believe that anyone can enter this industry, regardless of their background, as long as they have the willingness to learn. This isn’t advanced mathematics or quantum physics requiring specialized skills. I think the most important thing is to get a basic understanding of linear algebra and calculus with a good instructor. You don’t need to memorize theorems with proofs or deal with complex mathematical constructs. The required foundation is quite accessible and intuitive.
The real challenge is that, due to the popularity of the field, there are now countless courses—both paid and free—making it difficult to find truly valuable ones. My advice is to choose something that feels right for you, and if a particular course doesn’t make sense, try another one. Once you grasp the basic mathematical principles, you can then dive deeper into practice, algorithms, and various models.
Tell us about your key projects at Zalando—I came across your work on demand forecasting using transformer-based models.
In my first team at Zalando—Pricing & Forecasting—I worked on sales and demand forecasting for the entire product range available on our platform.
For this model, we used a transformer architecture. It has the capability to process data for thousands of items simultaneously and identify relationships within sequential data of any length. For example, we can analyze data for an entire year, and the model will determine in which months sales for specific products will increase or decrease.
At that time, in 2019, almost no one was using transformer models for such tasks. Typically, they were applied to text-related problems, such as translations, question answering, and document classification. Zalando was one of the first companies to apply transformers to time-series forecasting. In fact, academic literature on this application only started emerging in 2020—several months after our implementation.
The new model allowed the pricing team to calculate optimal discounts 15 times faster and publish them on our platform. Additionally, forecasting accuracy improved—the root mean square error (RMSE) of predictions decreased by 20.5%. The model also helped the company optimize stock replenishment. Customers no longer faced frustration when their favorite items were out of stock, and at the same time, the company reduced excess inventory of less popular products, improving warehouse efficiency.
Based on this project, we published a research paper. It was a groundbreaking achievement for the market, and in this publication, we aimed to clearly and comprehensively describe all our developments.
Now, I am working in the recommendations team, where we personalize content for users on the homepage. Our team worked on a project to recommend content to new users by analyzing their activity—looking at which products they browse, add to wish lists, and how they interact with the website overall. We discovered that these behavioral indicators were more significant for predicting interests than direct interactions with the homepage itself.
Using this data, we fundamentally redesigned our content recommendation model, leading to a significant increase in click-through rates and advertising revenue. Users started engaging more with sponsored content while receiving more relevant recommendations.
To summarize, how do you and your team contribute to the business?
Recommendation systems are the backbone of any online platform, whether it sells clothing, music, or video content. Recommendations are the key driver encouraging customers to make more purchases. While revenue per transaction can be increased through price adjustments, this approach has limitations, as each product has its own price constraints based on market competition. In contrast, recommendations and cross-selling significantly boost company revenue.
All recommendation systems are built on machine learning. The model studies a customer’s profile—demographic data, geolocation, device type, and browsing history. Based on this information, it predicts additional products the user might be interested in. Often, customers may not even realize they want a particular item until it is suggested. The goal of a recommendation system is to accurately anticipate user interests and present them with highly relevant products, ultimately driving significant revenue growth.
In which industries do you see the most active development of machine learning, and what new technological trend in ML seems the most promising to you?
The main trend right now is large language models. This is what all big tech companies are focusing on—OpenAI with ChatGPT, Google with Gemini, Anthropic with Claude are some of the most popular examples. The internet has accumulated such a vast amount of diverse data that it is now possible to train so-called foundational models—models designed not for specific tasks but for general use. These models can handle a wide range of tasks, such as translating text, identifying bugs in code, or even generating images and videos based on textual descriptions.
A similar shift is happening in time-series forecasting. For instance, there is a foundational model called TimeGPT developed by Nixtla, which has been trained on all available time-series data, including economic indicators, financial performance of public companies, weather data, and other publicly accessible information. You can upload data from an online store, for example, and the model will predict sales for the next quarter based on historical sources and existing patterns.
ML models are widely applied in medicine. Right now, new drugs are being actively developed, and ML helps predict which molecular compounds will be effective and which will not—this allows researchers to avoid long and costly studies. Machine learning significantly accelerates fundamental research in general: what used to take ten years can now be accomplished in just one.
Another key area is medical imaging diagnostics, such as MRI and X-ray scans. ML models are already outperforming human specialists in these tasks because medical and research institutions have accumulated vast amounts of data. As far as I know, many clinics intentionally share scans in public databases so they can be used to train models. For example, some models can analyze brain scans to predict conditions such as depression, Parkinson’s disease, and Alzheimer’s, providing recommendations for preventive measures to slow disease progression.
Robotics is also developing rapidly. While neural networks can now write coherent texts and create stunning images, household chores still fall entirely on human shoulders. Ideally, we want robots to finally take over routine tasks—like unloading the dishwasher or ironing clothes.
You mentioned that you support women in IT. What initiatives are particularly close to your heart, and what should change in the industry to encourage more women to pursue careers in this field?
I am actively involved in mentorship—for example, I lead a program on the online platform Women in Tech Global. I believe that female mentorship is especially meaningful because the IT industry is predominantly male, making it challenging for women to integrate into the community. For instance, men do not experience childbirth and are far less likely to take extended parental leave. As a result, they do not face multi-year career breaks that require rebuilding everything from scratch.
Women returning to the office after maternity leave often do not get enough time to reintegrate into workflows. During their absence, both the company and the industry may have changed significantly. Additionally, working mothers may not be entrusted with major projects due to assumptions that their attention will be divided between work and childcare, such as dealing with a sick child. Meanwhile, such assumptions are rarely made about working fathers.
More broadly, women often have less confidence in their abilities. For example, men tend to apply for jobs even if they do not meet all the qualifications and are less likely to struggle with impostor syndrome.
I wanted to share my experience of building a career after maternity leave and show that it is possible to grow significantly in one’s profession, even with multiple children. I also wanted to instill confidence in women, offer them support, and demonstrate that while taking on difficult challenges may be intimidating at first, it ultimately fosters invaluable self-belief and drives professional growth.
How do you manage to balance work, continuous learning, and personal life while working in ML, which requires constant development and surely takes up a lot of time?
First of all, my husband and I share all our daily responsibilities equally—we are both partners in a major project called “a family with four children.” Secondly, we try to delegate routine tasks to professionals—for example, we have a cleaning service, and three evenings a week, we have a babysitter who helps with school and kindergarten pickups.
For those who want to manage everything efficiently, I recommend carefully analyzing how you spend your time. For instance, if a meeting takes an hour and a half of travel time round trip but does not provide any new insights, is it really worth attending? When balancing a demanding job and a family, you often have to decline some meetings and less critical tasks.
Also, taking time to recharge and restore my energy is essential for me. My foundation is regular physical activity and good, healthy sleep. I find it much better to go to bed at 10 PM, wake up at 6 AM, and get things done in the morning rather than staying up late. I used to be a night owl, especially when struggling with a complex problem, but I would wake up feeling sluggish. Now, I have completely changed this habit.
Exercise also plays a crucial role. After intense intellectual work, it is important to switch to physical activity to relieve mental strain. We love going to the forest every weekend, and I need at least one full day in nature to reset from work.
I also love traveling. Many people say that changing the environment boosts productivity, and I can confirm that from my own experience. Every summer, we take a long vacation—for example, this year, we traveled with our four kids around Norway for a month in a camper van. Taking time to completely disconnect and travel with my family at least once a year helps me reset and regain focus.
