How a girl from Dushanbe created an AI startup for the government: the story of epsilon3.ai

Nadzima Noyoftova, city — Dushanbe, Executive Director at A7σ, Co-founder and COO at epsilon3.ai, Instagram

How it all started

I am the Executive Director of A7σ, A7Sigma, a holding company that unites AI startups in Tajikistan. Our holding includes several projects, including zypl.ai and zehn.ai.  

In 2024, our R&D team at zypl.ai completed the development of zGAN — a synthetic data generator — and handed it over to zehn.ai for further development, specifically for working on multimodality. We agreed that zypl.ai would receive a full license to use zGAN in FinTech.  

Since zehn.ai focuses on deep R&D, it does not engage in active sales. However, we have already seen applications of zGAN in other sectors.  

At the same time, we received requests for applying AI in the public sector. Given this interest, we decided to create the startup epsilon3.ai, which will focus on analytics and forecasting for the public sector and will receive a license to use zGAN.

Startup Launch

In September 2024, we began actively working on the startup, assembling a team from our colleagues in the holding. From the very beginning, three people worked on the project.  

In the first five months, we found partners in six countries: Tajikistan, Kazakhstan, Uzbekistan, Indonesia, the UAE, and Saudi Arabia. The rapid search for use cases is explained by the fact that government structures face similar problems to a large extent — the lack of quality forecasting, which leads to significant operational costs.  

To prove the effectiveness of our approach, the first case was forecasting housing demand in Kazakhstan. We used open data from egov on those in need of housing, supplemented it with regional macroeconomic indicators and synthetic data. As a result, the model trained on synthetic data showed 30% more accurate results than on historical data.  

We presented these results to various government structures in several countries, and they generated high interest.

Target Audience 

The main users of our solution are considered to be government structures that require analytics and forecasting. For us, it is not fundamentally important which specific use case to implement — forecasting electricity generation and demand or planning smart cities. The mechanics of operation remain unchanged; the only important thing is the presence of tabular data.  

Our solution significantly simplifies resource planning processes. Previously, clients conducted forecasting manually in Excel, using only historical data, which led to high inaccuracy. Now, in addition to historical data, the model includes macroeconomic indicators, which can have a significant impact, as well as synthetic data modeling various sudden events. This makes the model more robust and increases the accuracy of forecasts.  

At the moment, we are already working with JSC NIT and Karabatan Utility Solutions in Kazakhstan, as well as with several ministries and government agencies in Tajikistan, Uzbekistan, Indonesia, Saudi Arabia, and the UAE.  

Our solution has a direct impact on people's lives — thanks to more accurate forecasting, government structures can provide quality services, improving the comfort and quality of life of citizens.

Incubation Programs

Since we have existed for a relatively short time, we have not yet had the opportunity to participate in long-term acceleration programs. In December 2024, we participated in the Industrial AI Accelerator, organized by the Branch of the Autonomous Cluster Fund "Park of Innovative Technologies" in the city of Almaty, supported by the Committee on Artificial Intelligence of the Ministry of Digital Development and Aerospace Industry of the Republic of Kazakhstan. The accelerator was created to accelerate the implementation of AI solutions in key industries of Kazakhstan, connecting startups and developers with customers from the public and industrial sectors. The program helped find and test promising technologies, launch pilot projects, and integrate artificial intelligence into the business processes of large enterprises and government structures. A total of 30 startups were selected for the accelerator, and epsilon3.ai turned out to be one of seven startups that, at the end of the accelerator, signed a pilot project passport and implemented their pilot.

Difficulties

Providing services for the public sector inevitably involves encountering the fact that processes with them are not as simple as with the commercial sector. Here, decision-making processes are longer and more complex. For us, this is currently the main difficulty. Although despite this, we are already collaborating with government agencies in six countries.

Additionally, startups working in the field of AI for the public sector often have to overcome trust barriers. Government structures require proven reliability of technologies, especially when it comes to critically important processes such as forecasting, cybersecurity, or automation of administrative procedures. We solve this problem through pilot projects, which allow us to practically prove the effectiveness of our solutions.

Advice for startups working in this direction — start with small pilots, seek support from local partners who already have experience working with the public sector, and pay special attention to the legal refinement of contracts. This will help accelerate the implementation process and minimize potential risks.

Achievements 

In five months of operation, our startup has achieved significant results. We managed to establish partnerships in six countries and successfully implement several pilot projects. Within the framework of the Industrial AI Accelerator, we entered the number of six startups out of 30 that managed to bring the pilot to successful completion.  

One of our most significant achievements was winning the Industrial Startup Battle at Digital Almaty in the nomination "Most Revolutionary Solution," which confirms the innovativeness and prospects of our developments.

Plans

Currently, we are attracting investments in the amount of $300,000 with a company valuation of $6 million. Our plans include participation in leading international accelerators and entering new markets, expanding the geography of our solutions.  

The key product priority for this year is completing the work on zGAN multimodality, which will allow our model to process not only tabular data but also images and text. This will significantly expand the possibilities of its application and strengthen competitive advantages.