The streaming app Hera, developed by Kazakhstani creators, is already operational in the USA and European countries. Meanwhile, the mobile studio Defuse, from the same developers, is opening new opportunities for business and education. Both projects from Higgsfield AI have made it into the top 25 startups in the world. To discuss the development of IT in Kazakhstan, the creation and promotion of unique projects, and the use of AI, a correspondent from the Kapital.kz business information center spoke with the CEO and founder of Higgsfield AI, Yerzat Dulat.
- Yerzat Kanatuly, at the extended government meeting, you mentioned that in 9 months, you launched two products that have already gained international recognition, allowing you to enter the top 25 startups in the world according to Google. These are the mobile studio for content creation Defuse, which simplifies video creation from text, and a platform for generating content that opens new opportunities for business and education. What is the name of the second project?
- Defuse was our first application, launched in December 2023, which we actively worked on throughout the winter of 2024, and by spring, it peaked – reaching up to 2 million users.
After that, we began developing a new application, Hera. This streaming app will feature AI-generated content instead of real content. We aim to build a significant business.
Currently, we have not yet had a large-scale launch of Hera, but the app is already available in the USA and Europe, though not all content there is generated by AI; it includes some regular content as well. We are measuring how viewership statistics differ between AI and non-AI content, and we currently have data indicating that people cannot distinguish between the two. They watch both types of content equally.
Unlike older streaming applications where people primarily watch on televisions, Gera is designed for vertical videos. They are very short but super dynamic, as viewer behavior and content consumption have changed in the TikTok era. To be market-successful, content must be more TikTok-like. The primary audience for the Gera app is female.
We only recently started, about 2.5 months ago, and it's clear that we are just at the beginning of our journey. However, we are growing rapidly, with about 50,000 unique users watching us each month.
- How did the idea for these two directions come about?
- Initially, we were focused on DeepTech. We created a company that was one of the first in the world to train neural networks for video. We have been actively working in this direction for a year and a half and are still technologically among the top leaders in video generation technology. In the first 9 months, we had two directions: the first was training neural networks for video and media generation, as well as for images.
The second direction was that we needed to learn how to create quality consumer applications that could scale to a large number of users. This was the Defuse proposition, on which we built the infrastructure. It served as an intermediate application, and now the next stage is the development of the Gera app.
- What financial plans do you have for 2025?
- We are a startup, and everything changes quickly; we adapt rapidly, and there is a lot of strong competition in the market. So, the most important thing for us is to move quickly and be flexible. Globally, our plans for this year are to increase the number of users on Gera, grow the number of paying users, and continue to be technological leaders because the competition is very serious, with many Chinese and American companies in our field.
Currently, we have assembled one of the strongest marketing teams in the CIS, and our marketing approach is entirely data-driven.
At Higgsfield AI, we are growing rapidly. Initially, we spent a lot of money on technology development. We developed the technology and secured many patents, which makes our company quite strong in terms of intellectual property. Our technical developments, infrastructure, and robust team have allowed us to start creating commercial products.
Looking at the company's profits, they are growing exponentially. The numbers are rising quickly, but we are not disclosing them yet.
- At the government meeting you attended, the Minister of Digitalization mentioned that a fund of funds for venture investments will be launched in Kazakhstan based on Astana Hub. He said the target size of the fund is already one billion dollars, with preliminary applications from investors totaling 130 million. Will your company participate in attracting funding from their fund, and how will this, in general, help develop the IT industry in Kazakhstan?
- I have heard about it, and the idea will certainly have a positive impact on the industry. I believe it will be an effective solution because many new, very good startups have emerged, and the problem has been the lack of substantial funding. There are early-stage startups that raise small checks to find product-market fit: launching initial products into the market, and then they need to scale, requiring an exit for someone to buy them. However, there has not been a market for these last two stages in Kazakhstan.
Among the clear, old examples is the company Chocolife. Essentially, it was very strong, engaged in food delivery, but then, due to the lack of a market to invest further in them, foreign players simply came in and took over the market.
Therefore, I hope this will positively impact the industry, allowing startups to continue to grow and expand.
- Have you ever received government support, and overall, what do we lack in Kazakhstan for the development of IT talents?
- Kazakhstan is a social state. I think, to some extent, everyone has received government support. For example, I graduated from a top state physics and mathematics school. All my engineers graduated from our top state universities. Therefore, if we look at such basic things, the government has supported by investing globally in science and high-tech education: physics, mathematics, IT, machine learning.
- In an interview, you mentioned that Higgsfield AI is the future competitor of Open AI, but specifically in the media and video generation sector. Can you explain what reinforcement learning means and what the overall functions of an agent in an environment trying to optimize a reward function are?
- These are standard concepts from machine learning. To train a neural network, an optimization method called Reinforcement Learning is used when you train the neural network.
Open AI was one of the first to start working on this, which is why they are so big now. They directly trained ChatGPT to ensure that its responses were appealing to users. A large number of people sat down, whom they hired, to provide rewards. Rewards were given based on whether the bot answered correctly or incorrectly, whether it was appropriate or inappropriate, toxic or non-toxic, useful or not useful. This system is called "reinforcement learning."
You may have heard of the Chinese model DeepSeek, which was also trained using reinforcement learning.
There are few companies in the world capable of performing reinforcement learning; I was one of the first specialists, and I have written many works in this field. Not many people know how artificial intelligence works "under the hood." Reinforcement learning is the most complex and crucial aspect of training a neural network.
For a long time, no one believed in reinforcement learning. For example, I was engaged in it not on a commercial basis, but simply as research, independently of anyone, and then it turned out that it actually works. Now, when we attracted investments from America, our years of experience significantly helped.
- After finishing school, you decided not to enroll in a university because you felt that Kazakhstani universities would not teach you anything new. Essentially, you became self-taught, and by 2018, your ideas were being discussed and even used at Harvard and other universities and leading research labs. Moreover, you received a personal message from OpenAI's CEO Sam Altman, which you declined. Was there not a moment of hesitation during all this?
- I was very young; it was just youthful maximalism. Of course, my parents did not fully support this, as my grandfather was one of the first to receive higher education in Orenburg, Moscow, and St. Petersburg, and he was understandably quite worried about his grandson deciding