Najima Noyoftova is from Dushanbe. Initially, she didn’t plan to pursue a career in tech and earned a degree in sociology. However, after graduation, she began working as a recruiter at an IT company and soon realized she wanted to be more involved in the industry. She joined the Tajik startup zypl.ai, where she rose from CEO assistant to Director of the Executive Office. Since late 2024, she has been developing her own project, epsilon3.ai, which builds forecasting models and works primarily with the GovTech sector.
As part of the joint project by Digital Business and Astana Hub, «100 Startup Stories of Central Asia», Najima shared why she was initially hesitant to enter the IT field, how epsilon3.ai came to life, and how the solution can help prevent large-scale electricity theft. She also spoke about the challenges of working with government agencies to implement such a product , and why, for now, the project isn’t considering European countries for expansion.
«I used to think that IT was only for men»
– Najima, how did your journey in IT begin?
– In 2018, I graduated from school and had to decide which university to apply to. I was in a math-focused class, but I was convinced that IT was only for men and that it had nothing to do with me. So I chose to study sociology and enrolled at the Russian State Social University in Moscow.
While I was studying, I always had part-time jobs, and in my fourth year, I decided to look for a full-time position. I ended up working as an IT recruiter at an agency. Most of the time, I was hiring frontend and backend developers, but I also worked with ML engineers, AI product managers, and other specialists in the field of artificial intelligence. At first, I was still convinced that AI wasn’t for me, but at the same time, I realized I wanted to be on that side of IT.
After graduation, I returned to Tajikistan and started working remotely. But at some point, I realized it was time for a change. I began taking courses in product management and started looking for an offline position in Dushanbe, since remote work didn’t really suit me.
– Were there any specific careers you had in mind back then?
– I applied to many different positions, but deep down, I felt like I belonged in an international organization. Then, a close friend who had experience working with the zypl.ai team suggested I apply there. I picked a random position at the company and sent in my CV.
I was contacted by their HR team. I explained that I didn’t have much direct experience in IT, but I was truly eager to be part of their team. In the end, they offered me two options: office manager or assistant to the CEO. Of course, I chose the second one, and for about a year, I worked as Azizjon Azimi’s assistant.
During that time, I learned how a startup really works from the inside and significantly upgraded my skills. So, in the spring of last year, when the A7Sigma holding was just beginning to take shape, I was appointed manager of the executive office at A7σ. Later, I was promoted to director. As we continued working, our team began developing the idea that eventually became epsilon3.ai, a project initiated by the A7σ team.
«Our models achieve up to 99% accuracy»
– What does epsilon3.ai offer today?
– epsilon3.ai focuses on analytics and forecasting for GovTech companies, using zGAN — a synthetic data generator developed by the R&D team at zypl.ai. This technology is part of the GAN (Generative Adversarial Network) family, which is based on principles of game theory. A GAN consists of two components: the generator, which creates synthetic data that closely resembles real data, and the discriminator, which learns to distinguish between real and generated data.
What sets zGAN apart from other similar networks is its ability to specifically generate outliers — anomalous data points in the training set that differ significantly from the rest of the data. Outliers are typically removed in machine learning, as they can distort the model. We take the opposite approach: by using synthetic data, we add outliers to help identify deviations from the norm. For example, in the case of an energy company, zGAN can show where unexpected events might occur, such as a sudden spike or drop in electricity usage, or consumption reported with zero meter activity.
In one of our cases, we analyzed the market in Uzbekistan. According to official data for 2024, electricity theft in the country amounted to $108 million. If our model can detect even 20–30% of that, it could help save tens of millions of dollars.
– It seems like epsilon3.ai and zypl.ai are built on similar technology. So why was there a need to launch a separate startup?
– Originally, zGAN was developed for FinTech use cases. But as the technology evolved, it became clear that its potential could be valuable in other sectors as well, including government digital services. However, zypl.ai is focused specifically on FinTech solutions and wasn’t positioned to actively scale in the GovTech space. That’s why we decided to launch a separate project to explore opportunities in new industries.
– How did GovTech companies respond to your project?
– When we reach out to cold clients, we often face strong resistance. That’s why we chose a different strategy and started actively using the opportunities offered by the IT ecosystem across Central Asia.
At the end of 2024, we applied to the Industrial AI Accelerator, held with the support of the Ministry of Artificial Intelligence and Digital Development of the Republic of Kazakhstan. The program brings together businesses and government bodies to develop joint projects. That’s where we found our first client — the Kazakh company Karabatan Utility Solutions, for which we developed a pilot project.
After that, other clients began reaching out to us with specific requests. Many of our potential clients were already familiar with zypl.ai, which helped build trust. The company’s proven expertise in artificial intelligence and its international experience in FinTech gave us a strong foundation of credibility.
Astana Hub has also been a great support for us. Over the past few months, with their help, we’ve had the opportunity to present the project to the President of Kazakhstan, the King of Jordan, and the Mayor of Dushanbe. Since I’m based in Astana, I often receive direct inquiries. For example, we were recently invited to a meeting at the Ministry of Energy of Kazakhstan.
– Who are your potential clients beyond government structures?
– We also work with representatives of the quasi-government sector. One of our first pilot projects was with Karabatan Utility Solutions, an energy company in Kazakhstan that generates and sells electricity.
Our models reach up to 99% accuracy and can be applied beyond the energy sector. For example, forecasting for Smart City planning is an ideal use case for us.
In addition, we are actively developing our B2B direction. Our clients can be from any industry that needs to optimize operational processes. For example, we’re currently running pilot projects with telecom operators.
«If you want to work with the government, be prepared to do it for free»
– What kind of data do you need from the client to get started?
– It all depends on the client’s request. For example, if the goal is to forecast electricity consumption in a specific city district, we need detailed data — ideally, individual consumption records for each user or apartment over at least one year, and preferably for three years. Daily data is helpful, but hourly data provides the best results.
In one case, we received hourly consumption data for 15,000 users over the course of a year. Some data gaps were filled with synthetic figures, and the model still achieved 99% accuracy.
– Given the sensitivity of this data, how do you encourage clients to share it?
– All the data we work with is fully depersonalized. We don’t require any personal details like names or contact information. That said, we still take every precaution on our side to ensure there’s no risk of data leakage, even in this anonymized format.
– Representatives of government bodies are sometimes cautious toward young founders — especially women. Have you ever faced discrimination based on your gender or age?
– Fortunately, throughout my time in the startup industry, I haven’t faced gender discrimination. However, I have experienced some ageism. I often had to prove my expertise simply because of my age. I understand the hesitation — it’s not easy for clients to entrust important tasks to a young team.
That said, in the startup world, most founders are around 30 or even younger. Plus, Azizjon is a young entrepreneur who has already achieved significant results, and many people trust us because we are part of his team.
– What other challenges have you faced in your work so far?
– The biggest challenges have been bureaucracy, too many decision-makers, and, as a result, long negotiation and contract approval processes. The hardest part is often getting access to the person who actually makes the final decision.
At one point, someone told us, «If you want to do something for the government, do it for free». We took that advice and began creating free pilot projects for potential clients to demonstrate our expertise. Only after proving our value would we move into commercial cooperation. Now, we’re moving away from that strategy. We’ve built up a strong portfolio of successful cases, and we’re confident charging for our work from the very beginning.
I also often catch myself going through emotional swings — I’m either incredibly motivated, or I feel like nothing is going to work out. Sometimes that shift happens several times in just one week. Recently, we took part in the Silkway Accelerator organized by Astana Hub and Google for Startups, and the trackers told us that these ups and downs are completely normal.
During the accelerator, the project went through many positive changes. At first, we positioned ourselves as a startup focused solely on building forecasting models. But over time, we realized we had actually grown into a full-fledged AI lab, one that handles the entire cycle, from diagnostics and data analysis to developing and implementing tailored solutions.
– Does epsilon3.ai have competitors?
– In Central Asia, there’s a growing trend among startups to enter the GovTech space with a variety of products. Some may offer similar solutions, but the zGAN technology is unique to us.
epsilon3.ai was initially modeled after Palantir Technologies (the American company known for its advanced data analysis software for enterprises – note by Digital Business). Today, Palantir is both a source of inspiration and a competitor, although they don’t currently operate in Central Asia.
«This amount of work simply can’t be priced at $200»
–How many clients do you have so far?
– At the moment, we don’t have signed contracts, as the deal cycle in GovTech tends to be quite long. However, we only agree to develop a pilot project if it’s going to be paid.
– What is your monetization strategy moving forward?
– We’re working to help the GovTech sector get used to the SaaS model. Typically, we follow a standard approach: we develop the model and interface, provide all necessary dashboards, and, if needed, continue updating and improving the solution as new data becomes available.
That said, we’re flexible and adapt to our clients’ needs. Some organizations can’t work in this format due to internal regulations. For example, with one of our clients, we’re discussing an end-to-end solution, where we build the entire product from scratch and then provide ongoing technical support.
– What do you expect the average subscription fee to be?
– Each case is unique, and the final price depends on the scope and complexity of the tasks involved. That said, even the pilot stage typically costs several thousand dollars, because we carry out fundamental work for every client. In many cases, we receive raw data that requires significant preprocessing — cleansing, structuring, and preparing it for use. From there, we build a model from scratch, tailor it to the client’s specific goals, and develop a custom interface. A project of this scale simply cannot be priced at $200.
«We raised $175,000 at a $5 million valuation»
– How did you finance the development of the project?
– In the beginning, we operated with support from the holding team’s own funds. Then, in March of this year, we raised our first investment round — $175,000 at a $5 million valuation. The Kazakh fund White Hill Capital and Singapore-based Battery Road believed in us and backed the project.
– How long did it take to close the investment deal?
– I don’t think it took much effort to convince the investors. They believed in us almost from the very beginning. Plus, the funds were already familiar with zypl.ai, which helped build trust. From the start of negotiations to the moment we received the investment, it took just a few months.
The main point of interest during the process was exploring which other industries zGAN could be applied to. There was also a fairly detailed due diligence process focused on the team.
– What did you spend the funding on?
– The funds were primarily used to grow the team. I now work alongside 10 talented professionals on the development of epsilon3.ai. We’ve hired strong specialists from different countries, which has helped us speed up project delivery. What used to take a month, we can now accomplish in just a few weeks.
Expanding the team also expanded our capabilities. Previously, we focused solely on forecasting models, but now we have team members who can develop AI agents, a service that many of our clients are actively requesting, and one we can now deliver.
– When are you planning your next funding round?
– We’re currently in the process of raising a pre-seed round of $400,000. We already have a couple of soft commitments from venture funds in Central Asia. The funds will be used to support business development and to open offices in Tajikistan and Kazakhstan.
– What are your development plans for the next year?
– Our main focus right now is Central Asia. We’ve developed a number of strong pilot projects and are now working on converting them into revenue. Our goal is to reach a specific earnings milestone that will allow us to raise the next investment round and then begin expanding into Middle Eastern markets.
We’re unlikely to enter Europe in the near future due to strict regulatory barriers around AI. However, we may explore other regions, but only after we’ve achieved our primary goals. It’s important for us to grow step by step and stay consistent in our approach.