Марія Бровінська AI Eng 27 January 2025, 08:53

“Compared to my previous experience with Matlab and C#, it was like magic.” AI and ML developer at DataArt on LLM opportunities, overconfidence in AI, and the potential for those who master it

Eight years ago, in 2017, DataArt established the AI Lab department, led by Yuriy Gubin, Chief Innovation Officer at DataArt, and Dmitry Baikov, AI Technical Director at DataArt. Back then, they strategically predicted the demand for artificial intelligence in the future, and the company facilitated the creation of the corresponding department.

In the AI Lab, specialists work on their own AI developments, which are further used by DataArt and provided to customers. The company did not look for external specialists to work in this department, but created career development opportunities for its specialists, since at that time there was a shortage of qualified AI specialists on the labor market.

DataArt offered training to all interested colleagues. Specialists who successfully completed it and showed high results began working on both the company’s internal and client projects.

As part of a new series of materials on the use of AI in Ukrainian IT companies, dev.ua spoke with AI and ML developer at DataArt Anastasia Nehoda, who currently professionally develops the AI direction in the company, and learned what was achieved thanks to working with artificial intelligence.

How long have you been working with AI? Why did you decide to develop in this direction? What was your background before you got involved in AI?

— I have been working as an AI and ML developer for over three years. Before that, I was doing research. My first experience with AI was while working on my dissertation, where I used a combination of neural networks and fuzzy logic to assess the quality of telecommunications services. I was amazed at how effectively the algorithm worked.

Later, at IBM’s Data Science bootcamp, I trained a neural network in Python for the first time.

Compared to my previous experience with Matlab and C#, it was like magic: writing a few lines of code and seeing the result right away!

This experience inspired me to seriously delve into AI research. I started actively learning Python, attending thematic events. At one of these events, organized by DataArt, a speaker from Data Scientist at Spotify recommended a course on machine learning from Stanford professor Andrew Ng.

From that moment on, I began to systematically study algorithms, remember higher mathematics, and finally realized that I wanted to build a career in the field of AI/ML.

What are your responsibilities? What tasks does the company set for you?

— Depending on the project, my tasks can vary significantly. The main work includes the following stages:

Data preparation: collecting, cleaning, and creating datasets from various sources. This stage often takes the most time, but the success of the entire work depends on it.
Model development: algorithm selection, model training, and model testing to achieve optimal results.
Optimization: selection of parameters to improve the accuracy of models.
Integration and monitoring: implementing models into real products and maintaining their effectiveness.

Interestingly, the most difficult stage can be any of these processes — it all depends on the specifics of the project. For example, sometimes finding and preparing quality data is more difficult than creating the model itself.

I work in a team with ML developers, data engineers, analysts, architects, and other specialists. Everyone does their part of the work, and thanks to joint efforts, solving complex tasks is much easier.

AI/ML encompasses many areas and tools, so no one can be an expert in everything. I like that you can always turn to colleagues for advice or share your knowledge. Such interaction helps both to solve problems effectively and to develop professionally.

Can you boast about the results of your work — which processes have been optimized thanks to AI? Do you have an estimate of the savings in resources — time, money, human capital — thanks to the implementation of AI?

— I can think of several projects in which I participated and whose results I can be proud of:

Automation of cellular data analysis.
The cloud-based solution significantly accelerated the analysis of microscopic cell images and reduced the impact of the human factor. Speed increased by 40%, and accuracy and efficiency were significantly improved.
Optimization of banking processes.
Thanks to the automation of query processing and the use of natural language processing (NLP) methods, the process became continuous, reduced human participation by 90%, and significantly reduced the number of errors.
Processing of medical articles.
We automated the extraction of important data from scientific publications, which significantly reduced the time of experts, which is very expensive, and significantly reduced costs.

What inspires me most is the moment when people who previously spent a lot of time on routine work see how an automated system takes over these tasks and are sincerely happy with the result.

What AI tools are currently actively used in the company and for what? Do you develop your own AI solutions (if so, what are they and what are they used for)?

— From my experience, I can say that today a lot of attention is focused on projects using large language models (LLM), such as ChatGPT, Gemini, Claude, Llama, etc. Their popularity is easily explained by their versatility — they can be used in almost any field. For example, I was recently involved in projects in the financial industry, fashion, sales, and these are just some of the industries where this tool can be used. Although I cannot disclose all the details of the projects implemented at DataArt, the scope of LLM use is really very wide. For example:

in finance, they help automate risk analysis, document verification, or transaction classification;
in medicine — they provide guidance in the diagnostic process, structure clinical data, or create recommendations based on disease histories;
in tourism — personalize offers and improve customer service through interactive chatbots;
in the creative field — generate texts, images, music or scripts.

One of the advantages of LLM is its ability to adapt to specific tasks. For example, models like Llama can be deployed locally, which provides control over sensitive data and reduces cloud computing costs.

However, it is important to remember that universal models are not always the best choice. They can produce incorrect results, especially with poorly prepared data, or be inferior to specialized solutions in narrow areas. Our task is to understand when and which model is more appropriate to use, and to convey this to the client. Fortunately, now clients themselves are increasingly realizing that classical models trained to solve their specific problem, or a combination of such models, can be more optimal than universal LLMs. Therefore, other areas, such as computer vision, recommender systems, forecasting and optimization, etc., remain relevant.

In your opinion, will AI be able to replace programmers? And creatives? How do you feel about the fact that a new profession is already emerging — prompt engineer?

— At this stage, language models, which we call AI, are still a tool, not a full-fledged intelligence. They can significantly help programmers — reduce development time, automate routine tasks, or suggest solutions to certain problems. However, they are not yet able to completely replace programmers. It is always necessary to check the results, because the risk of model errors remains.

In the field of creativity, AI is already showing itself as a powerful tool. It is able to generate ideas, combine existing concepts, create visualizations or texts. However, it does not create fundamentally new ideas — this requires human intuition, experience and imagination. Artificial intelligence is more likely an assistant that enhances human capabilities, rather than replacing them.

Regarding the emergence of new professions, such as a prompt engineer, this is a natural stage in the development of technology.

We can’t stop progress, but we can adapt to change. It’s important to learn to use new opportunities to your advantage, finding a balance between automation and creativity.

Prompt engineers are an example of how people adapt to new challenges and create in-demand specializations in a rapidly changing world.

Do you agree that AI is a threat to businesses and even humanity to some extent? How can this be prevented and the negative consequences avoided?

— I believe that AI in itself is not a threat, but there are risks associated with how we use it and what processes we delegate to it. It is an extremely powerful tool that can bring many benefits — from optimizing business to revolutionizing medicine and science. But it is important to remember that it works on the basis of data that may be biased, outdated or incomplete, and AI itself can make mistakes or even «hallucinate.»

One of the challenges is overconfidence in artificial intelligence. People are starting to rely on it to solve serious problems, including making decisions at the government or business level.

But where human lives are at stake (in medicine, justice, or social policy), the final word should remain with humans. AI should assist, not replace, human judgment.

Another important aspect is economic change. Automation can force companies that do not have time to adapt out of the market.

The development of large language models is closely linked to the availability of large computing resources, making it accessible only to a limited number of individuals or organizations.

To minimize these risks, transparency is needed. AI must be understandable—how it works, why it makes certain decisions. We need clear ethical standards, regulations, and oversight mechanisms, especially where there are security risks.

People need to understand how AI works so they don’t fear modern technologies, but also realize their limits. Businesses need to work together to make these tools accessible and useful for everyone. Artificial intelligence is not about fear or competition, but about opportunities that we must learn to use correctly.

Please advise those who want to study AI, what to study, what courses, books, and blogs to pay attention to.

— I would recommend starting with the same courses that I started with: Machine Learning and Deep Learning Specialization by Andrew Ng. They are available on the deeplearning.ai and Coursera portals. I also recommend checking out The Batch section on DeepLearning.AI — there you can follow news from the world of AI.

From books, I can recommend «Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow» by Aurélien Géron — it provides both theoretical knowledge and practice.

In addition, basic knowledge of programming, probability theory, statistics, and higher mathematics helped me a lot. And of course, practice is important. Work with real data. For example, the Kaggle platform is a great place to practice and develop skills. The most important thing is not to be afraid to start, even if it seems that you don’t know enough. Learning by doing is not only the most productive, but also the most interesting way to learn something new.

I also advise you to attend thematic events, even online, because it is an opportunity to ask questions, get valuable advice, and communicate with like-minded people.

Anonymous bought the domain “OGOpenAI” and redirected it to a Chinese AI lab

OpenAI launched the Operator AI agent. It turned out that a similar idea was presented to Altman by Ukrainian startup Awesomic back in the fall of 2024

2025 is called the year of AI agents. Are AI agents really a must-have for everyone and what AI assistants are the Ukrainian divisions of EPAM Intetics Levi9 P2H and Railsware working on?

Dreams come true: OpenAI has shown an AI agent called Operator that can do the work for humans. What it can do, what makes it special, and who is it already available to?