– Najima, how did your journey into IT begin?
– I graduated from high school in 2018 and had to choose where to study. Even though I went to a math-focused school, I was convinced that IT was only for men and there was no place for me there. So I enrolled in the Russian State Social University in Moscow, majoring in Sociology.
During my studies, I constantly took part-time jobs. In my fourth year, I decided to look for a full-time position and managed to get a job as an IT recruiter at an agency. I mostly interviewed frontend and backend developers, but I also met ML engineers, AI product managers, and other specialists working with artificial intelligence. Although I still believed the AI field wasn't for me, I realized I wanted to be on the “inside” of IT.
After university, I returned to Tajikistan and switched to remote work. At some point, I realized it was time for a change. I began taking product management courses and started searching for offline positions in Dushanbe — remote work just didn’t suit me.
– What roles were you considering?
– I applied to many different vacancies and thought my place might be in international organizations. My close friend had worked at zypl.ai and recommended I try applying there. I picked a random open position and submitted my resume.
HR reached out to me, and I told her openly that I lacked direct IT experience, but I really wanted to be part of that team. They eventually offered me two positions: office manager or CEO assistant. Naturally, I chose the second one and spent about a year working with Aziz Azimi.
During that time, I learned how a startup operates from the inside and significantly improved my skills. So when the A7Sigma holding company began forming last spring, I was appointed Executive Office Manager for A7σ. Later, I was promoted to Director. During this period, our team came up with the idea for epsilon3.ai, which the A7σ team began developing.
“Our models reach 99% accuracy”
– What does epsilon3.ai offer today?
– We build analytics and forecasting tools for GovTech companies using zGAN, a synthetic data generator developed by the R&D team at zypl.ai. This technology belongs to the GAN family (Generative Adversarial Networks), which are based on game theory. A GAN consists of two parts: the generator creates synthetic data that closely resembles real data, while the discriminator learns to distinguish real data from synthetic.
What makes zGAN unique compared to other networks is its ability to deliberately generate outliers — anomalous data points that differ significantly from the rest of the dataset. In traditional machine learning, such data points are usually removed because they can distort the model. We do the opposite: we add synthetic outliers to detect deviations from the norm. For example, in an energy company case, zGAN can show where sudden events may occur: a sharp spike in consumption, a drop, or usage when meters show zero.
For one of our cases, we studied the market in Uzbekistan. According to official data for 2024, the country lost $108 million worth of electricity to theft. Even if our system identifies just 20–30% of cases, it could save tens of millions of dollars.
More on Digitalbusiness.kz website.
– Najima, how did your journey into IT begin?
– I graduated from high school in 2018 and had to choose where to study. Even though I went to a math-focused school, I was convinced that IT was only for men and there was no place for me there. So I enrolled in the Russian State Social University in Moscow, majoring in Sociology.
During my studies, I constantly took part-time jobs. In my fourth year, I decided to look for a full-time position and managed to get a job as an IT recruiter at an agency. I mostly interviewed frontend and backend developers, but I also met ML engineers, AI product managers, and other specialists working with artificial intelligence. Although I still believed the AI field wasn't for me, I realized I wanted to be on the “inside” of IT.
After university, I returned to Tajikistan and switched to remote work. At some point, I realized it was time for a change. I began taking product management courses and started searching for offline positions in Dushanbe — remote work just didn’t suit me.
– What roles were you considering?
– I applied to many different vacancies and thought my place might be in international organizations. My close friend had worked at zypl.ai and recommended I try applying there. I picked a random open position and submitted my resume.
HR reached out to me, and I told her openly that I lacked direct IT experience, but I really wanted to be part of that team. They eventually offered me two positions: office manager or CEO assistant. Naturally, I chose the second one and spent about a year working with Aziz Azimi.
During that time, I learned how a startup operates from the inside and significantly improved my skills. So when the A7Sigma holding company began forming last spring, I was appointed Executive Office Manager for A7σ. Later, I was promoted to Director. During this period, our team came up with the idea for epsilon3.ai, which the A7σ team began developing.
“Our models reach 99% accuracy”
– What does epsilon3.ai offer today?
– We build analytics and forecasting tools for GovTech companies using zGAN, a synthetic data generator developed by the R&D team at zypl.ai. This technology belongs to the GAN family (Generative Adversarial Networks), which are based on game theory. A GAN consists of two parts: the generator creates synthetic data that closely resembles real data, while the discriminator learns to distinguish real data from synthetic.
What makes zGAN unique compared to other networks is its ability to deliberately generate outliers — anomalous data points that differ significantly from the rest of the dataset. In traditional machine learning, such data points are usually removed because they can distort the model. We do the opposite: we add synthetic outliers to detect deviations from the norm. For example, in an energy company case, zGAN can show where sudden events may occur: a sharp spike in consumption, a drop, or usage when meters show zero.
For one of our cases, we studied the market in Uzbekistan. According to official data for 2024, the country lost $108 million worth of electricity to theft. Even if our system identifies just 20–30% of cases, it could save tens of millions of dollars.
More on Digitalbusiness.kz website.
Share
