Home page / Businesses / Expertise The QazСode company, which grew out of the IT department of Beeline Kazakhstan, has been working on the market for a couple of years. But its density of events is high. In addition to strategic cooperation with Beeline, export contracts have appeared in the company’s portfolio. QazCode became a member of the consortium to create the national language model KazLLM. And now he is planning to develop national models for other markets. The company is also actively implementing generative artificial intelligence internally – in coding and processes. But at the same time, she does not believe that AI will replace “white” employees, and is ready to hire cool specialists “even tomorrow.” Having (finally!) caught up with QazCode CEO Alexey Sharavar between numerous flights, we talked about the transformations that have occurred in his and the team’s lives recently. About why QazQode fundamentally does not want to be an “IT gallery”, but focuses on its own solutions and products. And with the help of our interlocutor we looked “under the hood” of KazLLM, learning how exactly large language models will be used in business. — QazСode has been on the market for more than two years. How did they go? — The first year we had a preparatory year, we call it a technical sprint.
We started by gradually transferring people from Beeline to QazCode. They did this in small groups. We set up processes and individual systems. The first 15 people were transferred in March 2023. And then gradually until the end of November the rest were transferred. In total, more than 700 people switched to us from Beeline Kazakhstan. The task was not to sow such a thing that people from QazCode would think that they now have nothing to do with Beeline. Although we have created an independent legal entity, we remain in the Beeline Kazakhstan/VEON group of companies. It was also important that people from Beeline did not change their attitude towards us either. So that this connection, the “chemistry” between teams of different companies remains. Therefore, in the first year there was a lot of communication with employees: we answered all questions, fought fears. They explained: “Yes, we are moving, but nothing changes from the point of view of our relationship with Beeline. We are still one family, we work together. But at the same time, we will also expand certain areas that will deal with foreign markets.” An important indicator is that no one left us at the time of transition. And the guys’ working conditions remained the same, they certainly didn’t worsen. — Do Beeline no longer have its own IT specialists?
— There are more than 180 people left who are responsible for core issues of the telecom business. It is important that these specialists are located within the company. We are talking about the billing, ERP system and cybersecurity support team. — The first year you transferred the team, set up processes in the new company. What happened in the second year? — Starting in January 2024, we began to form our own sales team to work with external clients. We currently have 5 sales people and 1 marketer. Yes, not much yet, but over the past year we have already sold and completed 12 projects with customers from 7 countries. It is important to understand: we do not work as an outsourcing company that simply sells the man-hours of its employees. The body shop model, as it is affectionately called in the IT community, has less margin, and, frankly speaking, it would not be very interesting for me as a CEO to work in it. This kind of business is definitely not what we dreamed of when creating a new company. QazCode today makes and supplies its own products and platforms to customers. We call this solutions. The main areas where we have strong expertise are the telecom industry, finance and retail. Some of the projects I mentioned were for the VEON group of companies – businesses in Uzbekistan and Kyrgyzstan. There were also supplies for
I have group headquarters in Dubai and Amsterdam. So far, most of our projects are in telecom. But we have already made deliveries for other areas: software for a well-known chain of fast food restaurants, a project for the financial sector. — That is, do I understand correctly: everything that you did for Beeline, you continue to do. At the same time, is the goal to diversify among clients, enter new markets, and receive foreign exchange earnings? - Right. We remain Beeline's main digital partner. About 40 teams within QazCode work only for Beeline Kazakhstan. And at the same time, we try to commercialize our solutions and the expertise accumulated over the years in export markets. In this regard, by the way, we are well supported by Astana Hub. They actively invite us to their events on the topic of IT export and on trips abroad. I think we are also making our contribution to achieving the goal set by the President of Kazakhstan - to reach $1 billion in IT export revenue. And this goal resonates very much with our internal ideology: to make cool craft Kazakh code. And to ensure that it is used not only in Kazakhstan, but also in other countries. — How has your life changed in two years, after you left the role of a manager?
IT at Beeline moved to the position of CEO of QazCode? — I started actively selling and... living on airplanes. Over the past year I have taken 42 flights. Otherwise, no big changes. — Has anything changed in QazCode’s culture after it was spun off into a separate business? — A feature of Beeline’s culture has always been entrepreneurship - they sold everything and everyone. This has been preserved with us as well. Somehow new guys who came to our company said: “Why doesn’t QazCode go for an IPO someday? To do this, you need to create software for the international market.” I say it’s a great idea, but we have to start somewhere. So we started. From the first sales to foreign clients. Many people, not just me, started going on business trips abroad. Communication with people from different countries and cultures. Our guys take it well. And if we talk about changes, we now have more English in our communications in the company. We have not completely switched to it yet, but we are moving in this direction. The immediate goal is to carry out 30% of the development in the company using GenAI - What KPIs are you looking at now, what will be the main metrics for the success of QazCode as a new business? — There are a lot of metrics, but the most clear and simple is the growth of revenue, our market share. And also
e increase in the portfolio of external orders. We strive to ensure that in five years the share of external and internal projects is 50/50. The second is income per person. It differs greatly between outsourcing galleys and companies that develop their own products. In the outsourcing business, revenue per person is at the level of $60-80 thousand, in the grocery business – as a rule, more than $100 thousand. These are global data. We still have less than $100 thousand per person, but we are striving for this. Third, increasing team productivity through the implementation of GenAI. I’ll say right away: I don’t believe that artificial intelligence will completely replace developers. There have already been so many different things that should “kill development” - low codes, no codes. Every 2-3 years some technology appears that they say will replace programmers. But in reality, the development market is growing even more. Yes, it will change as GenAI develops. A new category is emerging - industrial developers or AI developers: it will include juniors with a certain tooling who will work like middles. They will essentially become AI operators. If we are talking about coding, then there are about 10 different roles - developers, devops, analysts, designer
s, etc. And in each of these roles, you can use GenAI tools to varying degrees, which will increase productivity. Let me give you some numbers: just in test groups, our development showed an increase in efficiency when using artificial intelligence from 20 to 30%. And we want to extrapolate that everywhere. Ideally, up to 50% of the code should be generated using AI. Yes, there are security and data protection issues that prevent you from working with all available tools. But the general trend is obvious. By the end of the year, I want 30% of the work done by GenAI in all roles. But this does not mean that 30% of the guys, and we have very skilled ones, will be replaced by artificial intelligence. They will simply be strengthened with its help. And this will make it possible to increase productivity without significantly increasing staff. We already have more than 750 people in the company. And together with DAR Tech, with whom we signed a consortium agreement last year, we are the largest organization in the country by number of IT employees. We are proud of this, but there are also pitfalls. As the number of people increases, communications and processes become more complicated. How did KazLLM appear and why have its own model when there is ChatGPT - In partnership with the Institute
From Smart Systems and Artificial Intelligence (ISSAI) of Nazarbayev University, Astana Hub and with the support of the Ministry of Digital Development, QazCode became a co-developer of KazLLM. Tell us how the national large language model was created? — For us, it all started about 3 years ago with the idea of making a good bot for Beeline that would respond to users. We started putting together a data set: without recording personal data, we recorded calls, transcribed them, translated them into text, then finalized them... We spent a lot of money, and at the end we received 2 billion tokens, units of text. And for a normal language model you need at least 70 billion. As a result, they made a prototype - Kaz-RoBERTA-conversational, posted it on Hugging Face - this is a platform for all code related to AI and LLM models. Three months later, several thousand people downloaded it. We were surprised and realized that someone needed it. Therefore, we started looking for where to get more data. The next important step is cooperation with the GSMA (global organization of mobile operators - Digital Business note) and the guys from AINA Alliance, who created the Catalan language model. They have a very cool academic approach and great support from the rulers
ity - the leadership of Catalonia is worried about the preservation of the Catalan language in the digital era. With them we exchanged knowledge and experience on how to create a national LLM, with which we came to our authorities. And we came at the right time. It turned out that at the state level they were also thinking about this. And we, at QazCode, had data, we could provide the hardware, plus we had the largest team of data scientists in Kazakhstan (50 people who deal only with ML & DL, and a total of 150 people who work with Big Data, Data Science and GenAI). It was early 2024. We signed the consortium in June. Its participants were the Institute of Smart Systems and Artificial Intelligence (ISSAI) of Nazarbayev University, ICRIAP (the ministry acted as the coordinator of the process, bringing all specialists together), Astana Hub, Beeline Kazakhstan and QazCode. The guys from ISSAI were developing language models in parallel with us. They had great developers and linguists. However, even more resources were needed, including people and equipment. We provided the entire infrastructure, gave our team, and downloaded all the data that we had. After 9 months we had a large language model that
Paradise includes more than 150 billion tokens. 20% better than similar open source LLMs (for example, ChatGPT). In Kazakh language. The model also supports Russian, English and Turkish. In December, it was presented to President Kassym-Jomart Tokayev. Today KazLLМ belongs to the state. We have access to it and can use it for commercial purposes because we were part of a consortium. Separately, I would like to note that our large language model received international recognition in the form of an award at the GSMA Foundry Excellence Award 2025 in the Artificial Intelligence category. — You said the model is 20% better than its analogues. What does this mean? — There are different numbers of benchmarks; the guys from ISSAI and I did a comparative analysis. We have a very large database of the Kazakh language. Accordingly, if you make the same query to GPT and KazLLM, then our model will give a more accurate and expanded answer, using more words that it knows. And the response speed will be higher. In other parameters, our model corresponds to analogues. — So you are like DeepSeek and even cooler? - In some moments - yes. But they have a multi-lingual model, there are a huge number of languages. Our model is trained
only in Kazakh, Russian, English and Turkish. If you ask her something in Spanish, she won't answer. We chose the languages spoken here and foreign languages that are popular in Kazakhstan. In the future, we can add others - for example, Uzbek, Kyrgyz, Azerbaijani. And this will be a full-fledged model of Turkic languages. How can KazLLM be used in business - How can KazLLM be used in some applied tasks in business? And let me clarify: can only QazCode and Beeline use the model for commercial purposes or is this available to other companies? — The model is open source, anyone can download and use it for non-commercial purposes. How are we going to work with it ourselves? Beeline plans to introduce AI Tutor into the Janymda superapp (formerly My Beeline), a system that helps students and schoolchildren improve their knowledge by automatically generating lessons and tests in specific subjects. Support for multiple languages, including Kazakh, English and Russian, allows you to adapt the solution to the needs of different users, including foreigners learning the language. This will help make learning more accessible, effective and open new horizons for education.
tive process. And speaking specifically about QazCode, we already have several requests for the creation of national LLMs. There is a huge demand for this now. At the moment, if we want the state to use ChatGPT, it needs to be additionally trained with national data. This is already a custom project. Open AI does not undertake this. Accordingly, you either use open public models or create your own. Previously, it seemed that only the world's largest technology giants could create their own, but we have shown that this is not so. We are currently discussing the creation of a national language model for Bangladesh. Representatives of the Ukrainian government also came to us - at the international exhibition MWC Barcelona 2025, where we represented KazLLM at two stands - QazCode and GSMA. Uzbekistan, Kyrgyzstan, Malaysia, Syria, Oman – there is also interest. I thought LLM was a B2B project, but it turned out to be primarily B2G. And for us this is a promising business. The second commercial direction for LLM is RAG (Retrieval-Augmented Generation - “generation augmented by search”; combines the search for relevant information with text generation using a language model, can be used in legal departments for analysis
for and search for the necessary information in corporate documents, in finance - automated analysis of reports and provision of recommendations and other departments of the company - approx. Digital Business) and agents for corporate clients who work on the GPT principle and answer user questions. There is a huge demand for this now. For example, we are planning to launch a common platform within our own group of companies that will completely replace ChatGPT. We pay Open AI for its secure version of chat $65 per user per month, and we have about 6 thousand of them. In addition, we will continue to work with KazLLM. It is a living organism for us. Hot vacancies in QazCode - Based on the results of 2025, what will be a great result for QazCode and for you personally? — We have focus regions where we would like to develop – now there are about 35 countries. Our goal by the end of this year is to make 30 deliveries to these countries. Let me remind you that last year there were 12 projects in 7 countries. It will be an important result for me if we sell more craft Kazakh code to the world. And we won’t just sell, but make our products even cooler, and they will help our customers solve their problems. And, of course, I believe that by the end of the year we will
We can allow other countries to receive their own national LLM models. This inspires me a lot. — Final question: given such rapid development, who are you hiring now? — The hottest vacancies are data scientists (seniors and middles). After we launched KazLLM and became more recognizable in the market, we have 200 applications from juniors per month. But we need middles and seniors. Because such complex tasks as LLM for other markets require strong guys with experience. We have a lot of projects, the workload is high. And we will take as many new cool guys as we can get. We are ready to hire cool data scientists and data developers tomorrow – at least 50 people. No joke. We also need guys who are involved in sales - the team here is not large yet, but we are now actively developing it. We are also looking for local partners in our focus markets who will represent and sell craft Kazakh code. For these roles, fluent English is a must. And don’t be afraid of 42 flights a year :)
