A manufacturer of smart trays had a problem that no amount of fresh produce could sweeten. Its reporting systems could not give people a timely view of inventory, production, and deliveries. Information existed across the business. Getting it together quickly enough to act was another matter. Vitali Likhadzed described the project in a public account of ITRex Group’s work: the team built an operational data hub on Microsoft Azure, joining logistics, finance, and production information.
It is an unglamorous place to begin a story about an AI executive. There is no talking robot. There are goods that can expire, reports that can arrive late, and people who need to know what is happening. The account describes automated data movement, shared reporting, and access controls. The intended benefit is refreshingly legible: employees get usable information while there is still time to do something with it.
The data arrived. The answers did not.
Likhadzed is CEO and co-founder of ITRex Group. His public commentary travels through artificial intelligence, data engineering, and automation, but it repeatedly comes back to the distance between having a technology and getting something useful from it. In the tray manufacturer’s case, he described reduced waste and faster reporting. He supplied no percentage for either. The particulars of the problem are enough to explain his interest.
A sales team needs one answer, production needs another, and finance has its own way of organizing the world. Connecting their information is technical work with a social consequence: colleagues can finally discuss the same business. In Likhadzed’s account, the data hub also created a foundation for predictive analytics and AI. Those ambitions followed the work of making ordinary reporting useful. Even the future needs someone to sort the cupboards.
A business intelligence apprenticeship
His route to this work ran through the Belarusian State University of Informatics and Radioelectronics, where he studied from 1997 to 2002 and earned a diploma with honours. His professional biography identifies systems engineering as his degree subject. The university years also included additional English studies and a visit to the United States through the Work & Travel program.
From January 2005 to September 2018, he worked at EPAM Systems. His roles included leadership of the Minsk business intelligence practice and, later, Director of Technology Solutions. His responsibilities encompassed presales, staffing, and global strategy for business intelligence. They placed the commercial and organizational questions beside the engineering ones: how to shape the work, find the people, and deliver it.
September 2018 marks the start of his CEO role at ITRex. The company dates its own founding to 2009. Keeping those dates separate makes his career easier to understand. His documented move into the chief executive position came after more than thirteen years at EPAM. Today, his professional profile places him in Aliso Viejo, California.
A newsletter with an extra growl
By 2021, Likhadzed was publishing a LinkedIn series called Emerrrrrging Tech in Simple Words. The spelling did a little of the entertaining; the last three words did the explaining. Editions appeared across the year, including a combined September/October issue published in November. The subjects included retail data systems, livestream shopping, and emerging interfaces.
The title offers a useful introduction to this part of his working life. A chief executive selling complex engineering services also has to explain what those services are for. A buyer needs enough understanding to ask a sensible question before signing up for a complicated answer. Plain words help make that conversation possible.
His writing has often been collaborative. He has shared bylines with innovation analysts Andrei Klubnikin and Nadejda Alkhaldi, with Yelena Lavrentyeva on data challenges, and with ITRex CTO Kirill Stashevsky. The range of collaborators matters to the portrait. His public explanations bring business analysis and technical leadership into the same discussion, rather as the projects themselves must.
The expensive part around the model
A February 2025 guide to implementing AI, which he co-authored, asks companies to examine their readiness before expanding their ambitions. Talent, software, infrastructure, data, and business objectives all enter the calculation. The question is what an organization can actually support. A promising idea still needs people who understand the task and a system capable of carrying it out.
His cost writing makes another distinction that a purchasing department might appreciate. In a May 2025 article, ITRex estimated at least $50,000 for the AI components of a custom AI minimum viable product. Supporting web, mobile, cloud, and embedded applications would be billed separately. The article also acknowledged smaller-scale approaches using existing models. The estimate described a particular kind of development, rather than a universal price of admission.
ITRex’s estimate for AI components of a custom MVP. Supporting applications cost extra; smaller-scale approaches can cost less.
The qualification is part of the story. Likhadzed’s co-authored work on generative AI costs also examines choices about models, customization, and deployment. Those choices determine what a business must build and continue to support. Asking about the model is a beginning. Asking what surrounds it gets the conversation closer to the eventual bill.
ITRex itself now describes a practice spanning AI, data, and embedded systems. It reports more than 600 delivered solutions and more than 200 clients. Those are company totals, covering the organization’s work across its history. For its CEO, they form the setting for a continuing implementation problem: each new engagement brings an existing organization with habits, systems, and expectations of its own.
Six months, two weeks, and a necessary caveat
In a May 2026 post, Likhadzed returned to the limits of choosing a general-purpose language model and hoping for the best. “There’s a ceiling to what prompt engineering can fix,” he wrote. He discussed stages of training and adaptation, including supervised fine-tuning and reinforcement learning with human feedback. A business has terminology and tasks that a broadly trained model may handle inconsistently.
He reported an ITRex portfolio example in which fine-tuned models and human-feedback training reduced sales onboarding from six months to two weeks. The figures describe one reported project outcome. They offer no promise that another company will achieve the same result. What makes the example useful here is the measure he chose: the time needed to prepare people for work.
That is a more concrete yardstick than how eloquently a chatbot answers a demonstration question. It connects technical adaptation with a recognizable business task. His account puts the effort after model selection into view: teaching a system enough about its particular job that the people using it can get on with theirs.
That concern had appeared in his earlier work on big data. A 2022 article co-authored with Lavrentyeva examined five obstacles, including fragmented data, weak coordination, skills shortages, poorly chosen problems, and insights that never reach daily operations. Its recommendations included clear responsibility for data and teams able to teach colleagues how to use it. The subjects were familiar business intelligence concerns. Their reappearance in his AI commentary gives the newer work a history: the technology has changed, while information still has to cross departmental boundaries and become useful to somebody.
The colleagues inside the flowchart
His published advice on enterprise automation gives equal room to the organization around the software. A March 2025 guide, co-authored with Alkhaldi, sets out three areas of preparation: governance, data and technology, and people. It recommends defining the problem, improving the process before automating it, and establishing measures of success. Employees need clear explanations and training as their responsibilities change.
That guidance allows for modest purchases as well as custom development. Standard needs may suit an existing product. More particular workflows may justify something tailored. A seller of engineering services could make every problem sound bespoke. Here, the advice makes room for the possibility that a company should buy something simpler and get back to work.
The human emphasis has an echo in colleagues’ descriptions of Likhadzed. In a public professional recommendation, Andrei Dolinau described a leader who sought solutions rather than blame and combined directness with politeness. That is one colleague’s assessment. It fits beside the published advice on preparing staff for change, where explaining the work is itself part of the work.
A little room for the odd idea
There is also space in Likhadzed’s advice for work whose outcome is uncertain. Discussing the generative AI value chain, he encouraged businesses to create a dedicated research function, even a small one, and to experiment with unfamiliar technologies. He recommended recruiting people with unconventional ideas, while setting a limit on the budget for experiments.
“Invest in attracting the right talent and pay attention to people whose ideas seem unconventional.”
Vitali Likhadzed, on emerging technology
That combination gives his approach some texture. There is curiosity, expressed as room to try an idea. There is a boundary, expressed as money the organization has agreed it can spend learning. The same discussion describes ITRex employees being paid to learn during working hours. Knowledge takes time; someone has to put that time in the plan.
The career, the newsletter, and the project accounts meet at that practical point. Likhadzed’s subjects have moved from business intelligence into generative AI, while the questions around them remain recognizable. Can colleagues understand the information? Can they use the tool? Has anyone considered the cost of keeping it useful? The tray manufacturer needed answers before the goods expired. An elegant demonstration would have made a pleasant diversion. The reports still had to arrive.