Artificial intelligence has been transforming key industries for quite some time, and energy is one of them. AI optimizes complex processes and automates routine tasks, leaving employees with more time for creative and strategic work. Its true power lies in processing large volumes of data, which makes it possible to build deep analytics and accurate forecasts — the foundation of machine learning (ML) models and intelligent digital assistants that work with large databases through Microsoft Copilot Studio.
In cooperation with DTEK, MODUS X is implementing advanced AI technologies that help turn data into effective management decisions, automate routine processes, and increase team productivity. From Microsoft Copilot Studio integration to machine learning model development, these tools are already shaping a new data-driven culture across a large energy company and taking DTEK’s operations to a new level.
MODUS X is DTEK’s IT partner responsible for developing and implementing digital solutions. MODUS X grew out of an internal IT department and today operates as an independent company developing innovative products, including in the field of artificial intelligence. It is MODUS X specialists who stand behind the solutions that are changing DTEK’s day-to-day operations at every level — from user-facing Copilot tools to advanced ML models.
SPEKA spoke with Valentyn Vyntu, Chief Data Officer at MODUS X, about the company’s strategic goals in implementing AI, ML, and Microsoft Copilot Studio at DTEK.
We also spoke with MODUS X experts — Oksana Borysiuk, AI Implementation Officer, and Vadym Chuzhda, Head of Data Science — to learn how machine learning helps improve production processes at DTEK, how Microsoft Copilot Studio optimizes and increases work efficiency, and what they see as the future of AI technologies at DTEK and around the world.
MODUS X’s Strategic Goals in Implementing ML and Copilot at DTEK
Today, the energy sector is undergoing an unprecedented transformation, and data and artificial intelligence are driving that change. Our strategic goal at MODUS X is to help DTEK secure a leadership position in the industry through the implementation of advanced AI solutions, machine learning, and Copilot tools. This is not just about technology — it is about changing the way business is managed.
Valentyn Vyntu highlighted the following key priorities for MODUS X:
- Business process optimization: using ML models for forecasting, automating routine tasks, and supporting better management decisions.
- Scaling expertise: providing tools that enable DTEK teams — both IT specialists and business users — to effectively use AI and Copilot in their day-to-day work.
- Building a data-driven culture: creating an environment where decisions are based on analytics and digital assistants free up time for strategic work and development.
- Creating an innovation community: developing an internal community where employees exchange experience, mentor colleagues, and together raise the company’s digital maturity.
Valentyn Vyntu emphasizes that at MODUS X, artificial intelligence is seen not only as a tool, but as a strategic resource for sustainable growth and improved competitiveness for DTEK. That is why the company invests in team expertise, systematizes knowledge, and ensures access to the most advanced technologies.
“I am convinced that the future of energy lies in data-driven approaches and responsible use of AI. This is the path that allows DTEK to remain flexible, innovative, and resilient in the face of today’s challenges.”
— Valentyn Vyntu, Chief Data Officer
Machine Learning Solutions at DTEK
Machine learning (ML) is an artificial intelligence technology that enables computers to learn from data and make decisions or predictions independently. Instead of following explicit instructions, the system analyzes examples, identifies patterns, and applies that knowledge to new situations.
Current Projects
We asked Vadym Chuzhda which processes at DTEK are optimized using ML. He explained that MODUS X has been maintaining a system for predicting wear and breakdowns in drilling equipment for quite some time. The system works with data from numerous sensors and helps identify potential risks in time.
Another case is automated parsing of insurance documents. This is an internal service used annually to collect insurance policies. The data from these documents is structured using ML algorithms, which makes it possible to analyze it efficiently and make informed decisions.
One of the most relevant and strategically important projects at the moment is coal distribution optimization, which helps the client improve operational efficiency.
MODUS X is also testing a Weather-Dependent Energy (WDE) forecasting system, which includes solar power plants (SPP) and wind power plants (WPP). The WDE solution is already undergoing pilot industrial testing at three sites and makes it possible to factor in weather conditions for more accurate generation planning.
Optimization is one of the most relevant solutions in digital business transformation today. Why? In part because of the popularity of large language models (LLMs), such as ChatGPT.
Users are increasingly turning to LLMs with their own requests. While these models cannot fully solve every problem, they can help identify and partially address tasks. This has drawn attention to optimization as a field: today, MODUS X is working on or launching around five projects aimed at improving different business processes. One of the most advanced and already nearly implemented projects is the coal distribution optimization system, which is being prepared for its second release.
DTEK has a network of mines and coal enrichment centers (CEC), because not all extracted raw material is immediately suitable for power generation. Coal from different seams requires different treatment: some of it can be sent directly to thermal power plants, while other types must first undergo enrichment.
The optimization task consists of solving a complex logistics problem: identifying the most efficient coal transportation routes — from mines to enrichment centers, then to stations or to external buyers such as heating plants. It is important to take into account the constraints of each stage: the daily processing capacity of the CECs, warehouse capacity, station demand, and commercial sales options.
This is practically impossible to solve manually, which is why a mathematical model is being developed to take all constraints and usage options into account. However, as Vadym Chuzhda notes, this is far from an academic exercise: in real business, new conditions arise every day — production volumes fluctuate, informal storage sites or wagons left on the tracks may appear, and logistics capabilities may change. All of this needs to be analyzed, described, and only then formalized in mathematical models. The process takes time — from six months to a year — and requires a deep understanding of the business’s internal logic.
“This is currently our most visible project, and the effectiveness of the solution is already clear. Other projects are now being aligned with it so that we can scale its success further.”
— Vadym Chuzhda, Head of Data Science
The Typical MODUS X Workflow for Implementing ML Solutions at DTEK
Work begins with business analysis. At this stage, the team works together with business analysts to clarify the essence of the task: what exactly the client wants to achieve, what constraints exist, and what data is available. Based on this, a preliminary technical specification is developed and the scale and duration of the project are estimated.
Next comes a deeper analysis of the available data — sometimes initial estimates need to be adjusted, because not every request can be implemented without the right information. If the data is unavailable, certain solutions simply cannot be built.
The next step is dataset preparation and the creation of baseline models. The team tests several variants without fine-tuning to determine which approaches are most suitable: in some cases, a time series model is the right choice, while in others a simpler classification approach is more appropriate. Model selection depends not only on accuracy, but also on additional requirements such as processing speed, task specificity, and infrastructure limitations.
After that, the selected models are refined through fine-tuning and a demo version is created, which can serve as a proof of concept. If the client provides the full dataset from the outset, this stage moves directly into implementation — the model is then integrated into the business process.
Deployment is handled by the DevOps team: the model can be integrated into internal resources or handed over to the client for independent use. For example, its forecasts can be sent to a database from which analytics is performed through Power BI.
After implementation, the team continues to support the model and monitor its performance and stability.
Data Preparation and Business Analysis Usually Require the Most Resources
This is a classic Data Science pattern: 80% of the time is typically spent on understanding and preparing the task, and only 20% on building the model. In complex cases, such as optimization models, the balance can be even more skewed: 90% of the effort goes into analysis and data preparation, and only 10% into the model itself.
“This is a classic situation in Data Science, where 80% of the time is spent understanding and preparing the task, and only 20% on building the model. In complex cases such as optimization models, this ratio can be even higher: 90% of the effort goes into analysis and data preparation, and only 10% into the model itself.”
— Vadym Chuzhda, Head of Data Science
In addition, the team faces extra difficulties because of the war: information may be lost due to shelling, or access to data from temporarily occupied territories may be unavailable. This significantly complicates the preparation of a complete dataset.
Trends in ML Forecasting and Optimization
Optimization itself is in growing demand. The expert notes that this may be partly related to the war — businesses are looking to save money, reduce costs, and improve efficiency. If five years ago people talked much less about business process optimization and the use of mathematical models, today such requests appear constantly. One project often leads to the next — and this momentum continues to grow.
“I think that with the development of large language models such as GPT, it makes sense to look at changing approaches to forecasting.”
— Vadym Chuzhda, Head of Data Science
Vadym Chuzhda sees the use of large language model architectures, especially transformers, for forecasting tasks as a promising direction. According to him, more and more models are emerging that allow time-series forecasting by analogy with text data. Such approaches include GPT-like architectures as well as specialized solutions such as Informer or PatchTST.
Vadym Chuzhda stressed that this is a new but very interesting direction that should be explored for potential use in large energy companies such as DTEK.
The Role of Copilot Studio at DTEK
DTEK is actively implementing AI solutions of varying complexity. One of the development directions is the use of Copilot Studio, a low-code platform that allows business users with minimal technical experience to independently create AI agents for their own needs. The latest project by MODUS X focuses specifically on user support: helping them create and effectively integrate Copilot Studio agents into business processes.
At the same time, the company has teams of experienced developers, including data scientists. They build more complex AI solutions that require deep technical expertise and use a broader set of tools, not limited to Copilot Studio.
At MODUS X, two parallel tracks are being developed for implementing artificial intelligence into DTEK’s operations:
- Simple AI solutions that employees create themselves for their own needs.
- More advanced solutions developed by the in-house MODUS X team — either independently or in partnership with specialized IT companies.
Types of Copilot
Users can adapt Copilot to their own needs — each team configures the tool according to the specifics of its work. This makes it possible to integrate AI more effectively into everyday tasks.
Oksana explains that there are currently three main types of Copilot used within the company:
- Copilot Chat — a basic tool available to all users with a Microsoft Office package. It is a chat assistant similar to ChatGPT, designed for conversational interaction.
- Microsoft 365 Copilot — more deeply integrated into Microsoft applications such as Word, Excel, Outlook, and others. However, its use depends on licensing, so not all employees have access to it.
- MS Copilot Studio — a low-code platform that enables teams to create their own AI agents for specific tasks. This tool opens up broad possibilities for customization and for building solutions that are as close as possible to business needs.
According to Oksana Borysiuk, MS Copilot Studio is actively used at DTEK in the following areas:
- Creating chatbots that operate based on internal team knowledge or external sources such as websites containing legal requirements, regulations, or industry standards. These sources are often combined — for example, to compare internal documentation with current regulations.
- Employee training. One of DTEK’s departments has created a chat agent based on internal knowledge that was previously shared as a set of files. Now employees can ask the chat any question instead of searching for information manually.
- Interactive testing. A chat tool with knowledge-testing functionality is currently in development. Employees answer questions, and the system immediately evaluates the responses, points out mistakes, and highlights what needs attention. This helps identify knowledge gaps and potentially retrain employees.
- Support with technical questions. Copilot Studio is used as a reminder tool. For example, instead of asking colleagues, “Do you remember how to do this?”, an employee can ask a chatbot that has access to the relevant knowledge base.
Перевага Copilot Studio у тому, що інтегрувати його дуже легко. У ДТЕК вже є продукти Microsoft, які є частинами Microsoft 365, тому не було потреби в якихось додаткових інтеграціях.
Водночас у процесі експлуатації виявилися певні труднощі, пов’язані зі стабільністю самого інструменту. Copilot Studio доволі нова платформа, яка активно оновлюється. Через це, наприклад, можуть змінюватися елементи інтерфейсу: налаштування, які працювали минулого разу, виглядатимуть інакше або потребуватимуть нових дій. Тому команда вирішила проводити навчання не у форматі разового курсу, а створила постійний Copilot Studio Club, де учасники можуть обмінюватися інсайтами та консультуватися з більш досвідченими колегами.
Часті зустрічі у форматі клубу та обмін думками допомагає цей інструмент [Copilot Studio] краще освоювати, натепер по ньому немає нормального навчання, адже він новий і розвивається дуже динамічно. — Оксана Борисюк, AI Implementation Officer
Клуб відкритий для всіх працівників ДТЕК, спеціалісти MODUS X є його кураторами та ідейним натхненниками.
Скільки часу знадобилося, щоб інтегрувати MS Copilot Studio у ДТЕК
The main advantage of Copilot Studio is how easy it is to integrate. DTEK already uses Microsoft products that are part of Microsoft 365, so no additional integration work was needed.
At the same time, some challenges emerged during operation, mainly related to the platform’s stability. Copilot Studio is a fairly new platform that is actively evolving. As a result, interface elements may change — settings that worked before may look different or require new steps. That is why the team decided not to run training as a one-time course, but instead created an ongoing Copilot Studio Club, where participants can exchange insights and consult more experienced colleagues.
“Frequent club meetings and the exchange of ideas help people master this tool [Copilot Studio] more effectively. Right now, there is no proper training for it because it is new and developing very dynamically.”
— Oksana Borysiuk, AI Implementation Officer
The club is open to all DTEK employees, and MODUS X specialists serve as its curators and driving force.
How long did it take to integrate MS Copilot Studio at DTEK
It is important to keep in mind that Copilot Studio is designed for quickly building simple chatbots. This leads to its main limitations:
- Context window size, meaning the amount of information the bot can work with at one time. For example, if a 600-page document is stored in SharePoint, MS Copilot Studio will not be able to process it in full. But if the same file is uploaded directly into the bot’s memory, the result will be significantly better. In other words, the format and method of connecting data strongly affect answer quality.
- The number and diversity of documents. If a bot has access to a SharePoint site with hundreds of files, especially on different topics, it can get lost in the volume and provide less accurate or less relevant answers. Oksana Borysiuk recommends structuring the information — for example, dividing data by topic or creating separate chats for each folder. This improves quality but requires more time and effort from the user. As an alternative, Microsoft Azure AI Foundry can be used.
- MS Copilot Studio currently does not allow model selection — it uses GPT-4o with no option to switch to other models.
If greater flexibility or the ability to work with large data volumes is needed, Oksana Borysiuk recommends looking at Azure AI Foundry.
“These two alternatives we are considering are from Microsoft, so they are the most accessible for us in terms of security and configuration.”
— Oksana Borysiuk, AI Implementation Officer
Positive changes at DTEK thanks to the integration of AI assistants into work processes
Oksana Borysiuk shared that introducing AI into daily workflows has had a positive impact on employee efficiency.
It is still difficult to measure the exact impact of Copilot Studio integration, because most of the solutions are relatively new, some are still under development, and active usage has only just begun. However, positive changes are already visible, especially in the time saved on routine tasks. If we look not only at chatbots based on Copilot but also at the internal training initiative “AI Education,” we can already see more efficient workflows and a greater focus on creative work.
“Colleagues have confirmed to me that now they are doing interesting tasks and analyzing information, rather than searching for it.”
— Oksana Borysiuk, AI Implementation Officer
Future innovations and directions for Copilot at DTEK
We asked Oksana Borysiuk about future implementations for DTEK and possible changes related to Copilot:
“There probably will not be major changes. We plan for logical development, and if users who have already mastered Copilot want to work with more advanced tools, we will fully support that.”
The expert emphasized the importance of systematizing knowledge and developing an internal community:
“Right now, our specialists [MODUS X] are the mentors. The next stage is when the first group of advanced users becomes mentors for their colleagues. That is when our AI wave will spread even faster, and MODUS will be the organizer and source of inspiration for this process.”
Trends in the development of AI assistants and MS Copilot Studio-based solutions for 2025 and beyond
Oksana Borysiuk highlights four key directions in the development of AI assistants built on Copilot Studio:
- Employee training — for both new and current employees. AI assistants can do more than answer requests: they can proactively suggest learning materials, track knowledge retention, and point out what should be studied next.
- Working with regulatory frameworks — including integrating chatbots with the websites of the Verkhovna Rada, tax authorities, and other official sources to work with current laws, standards, and regulations. This is one of the simplest and most useful use cases for Copilot Studio.
- Internal knowledge bases — chatbots that help employees quickly find the information they need without searching manually through documents or databases.
- Project status tracking — another important direction that makes it possible to monitor the progress of initiatives, including AI-related ones, more effectively.
