Ainur Nuradil from Almaty used to work at the Academy of Sciences. Then she came up with AR Sana, a mobile app that helps Kazakh-speaking kids develop their speech using AI and augmented reality. The team has already trained its own AI model on children’s voices and is now getting ready to launch the service.
For the joint Digital Business and Astana Hubproject “100 Startup Stories from Central Eurasia”, Ainur shared how she sold the idea to a speech therapy center before the product had even launched, and how she managed to pitch the project to AI and Digital Development Minister Zhaslan Madiyev in just a couple of minutes in an elevator. We also talked about why a ready-made AI model worth $70,000 didn’t work for the Kazakh startup, and how the team won more than 15 million tenge through various accelerator programs.
“A speech therapy center paid 89,000 tenge before the product even launched”
— Tell us about yourself. What were you doing before the startup?
— I’m from Almaty. Like everyone in my family, I graduated from a physics and math gymnasium. We have three generations of teachers in our family, including master’s degree holders, PhDs, doctors of science, and academics. After earning my bachelor’s degree in IT, I went to work at the Academy of Sciences, in the laboratory for analysis and modeling of information processes at the Institute of Information and Computational Technologies. My focus was text data analysis. Later, I started working with machine vision.
Later, I went to study Design & Media at New York University through the Bolashak program. When I returned to Kazakhstan, I started working as a senior lecturer at Astana IT University. What I enjoyed most was supervising students’ theses and projects. Gradually, I felt the urge to build something of my own. That’s how I came up with AR Sana, a mobile app that helps children develop speech through play.
— How did the idea come about?
— When I was studying in the US, my son had just turned three. At first, he only spoke Kazakh, but our fellow Kazakhstanis in New York mostly used Russian when talking to each other. Because of that, my child started forgetting his native language. When we returned home two years later, my son could no longer pronounce some of the more difficult Kazakh letters correctly. That’s when I wanted to create a platform for teaching children the state language.
I came to Astana Hub to test my product idea and show it to experts. That’s where I accidentally found the Startup Garage 2025 incubation program. I sent in my pitch deck and got accepted. We worked with trackers and dozens of experts there. Daniil Matskevich, co-founder of Squares.kz, an event venue rental platform, told me my project didn’t really look like a startup yet. Basically, if Duolingo added Kazakh, my idea would instantly lose its edge. So I started figuring out what a startup should actually be and began with in-depth interviews with parents.
During those conversations, I realized there was a much bigger issue: speech delays in children. Every fourth parent brought it up. That’s how the idea for a mobile platform came about, where preschoolers could do speech therapy exercises to develop their speech. First, we worked with speech therapists to build the methodology, and then started turning it into a gamified app. The idea was that a neural network would check whether the child was doing the exercises correctly.
– How did you get started?
– We started by researching the problem. I visited dozens of speech therapy centers to see whether there was real demand for an app like this. At one point, I walked into a center pretty much at random and sold the concept before the product even existed, for 89,000 tenge. The amount was random too. Back then, I hadn’t really thought through the financial side yet. That’s when I realized specialists were actually ready to pay for this kind of app.
I needed a team to build the project, so I decided to look for people on Instagram. I simply posted, “I have this idea. Who wants to join?” My thesis students from Astana IT University responded. They’re still working on the startup today. There were five of us at the beginning, and we’ve been moving forward together ever since. Galymbek, our CTO, is the lead developer behind the game mechanics. Nurai, our COO, handles operations. Samir is our 3D artist, and Bibimaryam combines audio design and SMM. We also have people working part-time and on an outsourced basis.
“I ran after the minister and jumped into the elevator to tell him about the project”
— How does the app help kids improve their speech?
— AR Sana is a platform built around two main components.
AR Sana App is a mobile app built as a series of games and exercises. A child looks at the phone screen and sees themselves and their room, while the app uses augmented reality to add virtual characters and interactive tasks.
For example, a character might demonstrate an exercise, like blowing at the screen or rolling the tongue into a tube shape, and the child repeats it in front of the camera. If the exercise is done correctly, the game moves on to the next step.
AI, speech recognition, and facial movement tracking systems check whether the exercises are being done correctly. We plan to have 350 games in total, and more than 80 are already ready.
AR Sana Lab is a web platform for speech therapists and specialists. They can track a child’s progress, see homework data through charts, and get ready-made reports. If a child cannot complete a certain level, the speech therapist can make it easier. One important point: we don’t position AR Sana as a replacement for a specialist. We only take on the homework part and serve as an extra tool.
The name has several meanings built into it. AR stands for the augmented reality technology we use. Sana means “intelligence” in Kazakh. And together, “ar sana” means honor and dignity. That’s what guides our team in our work: doing everything for good.
— At what age can children start using it?
— From about a year and a half old. At that age, children should already be starting to say words and short phrases, name objects, and use simple verbs like “give,” “show,” or “want.” If those skills are developing more slowly, it’s worth paying attention and consulting a specialist. For younger children, we focus on helping launch speech development, while for older kids the exercises are aimed at pronunciation and building connected speech.
– What do the lessons look like?
– The lessons last between 7 and 13 minutes and are built around game mechanics. Peer-to-peer learning is considered one of the most effective approaches, so in AR Sana, kids interact not with an adult teacher, but with a virtual girl their own age named Sana. We gave her a full backstory: her personality, the words she often uses, her habits, even her zodiac sign. She has leadership traits. She guides kids forward and isn’t afraid of challenges. Besides Sana, there are three other characters: a speech therapist, Amirtai, and Yesen.
— A neural network tracks how the child pronounces sounds. What model are you using?
— The existing Kazakh-language model from ISSAI, the Institute of Smart Systems and Artificial Intelligence at Nazarbayev University, was trained on adult voices, so we’re building our own AI model based on children’s voices with correct pronunciation. To collect the dataset, we worked directly with schools and parents.
Collecting audio recordings was moving slowly, but then a chance encounter helped. At one Astana Hub event, Kazakhstan’s Minister of AI and Digital Development, Zhaslan Madiyev, was there. After his speech, he was leaving, so I ran after him and jumped into the elevator. He was surprised, and so were the security guards. It turned into a literal elevator pitch with the minister. While we were going down to the parking lot, I managed to explain AR Sana’s technical problems and what kind of help we needed. After that, we had a more detailed conversation about the project. I told him about AR Sana’s technical challenges, the need to collect children’s voice data, and possible support from educational institutions. Zhaslan Khasenovich took the initiative seriously and said he would discuss it with the Ministry of Education.
Soon, educational and correctional centers started reaching out to us, and that’s when the project moved into a new stage. Thanks to our work with schools, speech therapists, and parents, we significantly sped up the collection of the speech dataset needed to train our AI models.
In just one month, we managed to collect more than 17,000 short audio recordings of children’s speech, all with official consent from parents and representatives of educational institutions. It was extremely important for us to follow all ethical and legal standards. The data was collected anonymously, without revealing the children’s identities.
For us, this wasn’t just about technology or building a dataset. It also showed that parents, specialists, and schools are ready to come together around the idea of early support for children and creating an accessible system for speech development.
Of course, we could have taken an easier route. In the US, a ready-made English speech recognition model costs anywhere from $50,000 to $70,000. But our goal is to launch the product in Kazakhstan, in the state language, and only then scale abroad.
— When will the app become publicly available?
— We’re currently at the pre-launch stage, with the release planned in about two to three months, around mid-summer. Right now, speech therapists from different centers are testing AR Sana. We’re already running pilots at Easy Study, LogoAzbuka, and several schools in Astana, Semey, and Almaty. At the same time, we’re gathering feedback and testing the practices that will become part of the first product version. The launch version will include around 150 game scenarios.
“For some parents, it’s very hard to admit that their child has a speech delay”
— How is the project going to make money?
— We’re using a B2B2C model, where speech therapists recommend AR Sana to parents. The free version of the app includes five basic exercises. Full access for parents will cost $9.99 per month.
We also offer a separate Enterprise subscription for institutions such as speech therapy centers, schools, and kindergartens. It costs $1,500 per year. In that case, access is provided not to a single parent, but to the organization itself. Specialists can use AR Sana while working with children, connect multiple students, track their progress, and assign homework to parents through the app.
— What are you doing to attract customers?
— We started by cold-calling centers, but it didn’t really work. We almost never even got invited in for a conversation. Then we realized the best way was to just show up in person and try to meet the people in charge.
Media helped a lot too. Khabar TV and Egemen Qazaqstan covered us, and after that, speech therapists saw we weren’t some kind of scam and started listening. Mentioning that AR Sana uses AI also helped. In one of his addresses, Kassym-Jomart Tokayev talked about the importance of developing this technology, so state institutions had already heard about it. When we presented the product and explained that it works with artificial intelligence, specialists became much more interested.
Right now, we’re working with 5 centers and 2 schools. In total, more than 35 specialists are using AR Sana.
— How do parents react when speech therapists recommend the app?
— Many parents understand how important regular independent practice and early work on a child’s speech really are. But in practice, there’s another much deeper issue: emotionally, it can be very hard to accept that your child may need help.
Because of fear, anxiety, or the hope that “they’ll grow out of it,” families often put off seeing specialists and lose precious time during those early developmental years. Parents are also often influenced by the older generation. Grandparents may reassure them by saying things like “all children are different,” “they’ll start talking later,” or “everything happens in its own time.”
Sometimes that’s true. But in some cases, early support and timely diagnosis can play a key role in a child’s further development. That’s why we try to talk about speech development gently, without shame or pressure on parents, so that getting help feels not like a verdict, but like caring for the child’s future.
Sometimes parents ask us, “How can a child work on speech through an app?” But we’re not saying AR Sana replaces a speech therapist or live communication. It’s an additional tool: the child works with a specialist, communicates with family, and the app helps them regularly repeat the exercises in a playful way and reinforce the results.
“I jumped into a pool and pitched the project in freezing cold water”
— How much money has gone into developing AR Sana?
— I invested my own money, around 7 million tenge. Most of the expenses went toward paying programmers and designers. At the same time, a lot of people were willing to help for free. For example, speech therapists helped collect voice samples to train the AI model.
We also take part in different accelerator programs, and altogether we’ve won around 15 million tenge through them.
— Which accelerator programs have you taken part in?
— Before coming to Astana Hub with my idea, I knew absolutely nothing about startups. We first joined the three-month Startup Garage incubator and ended up making the top 15 out of 308 projects. That’s when media outlets started paying attention to us. Later, we took second place at Sales Bootcamp, another three-month program by Astana Hub. After that, we joined the Social Startup program by NU Impact Foundation and made it into the top 10 out of 150 startups on Demo Day, where we won 10 million tenge.
Recently, we applied to the World Bank’s venture program, and just a few days ago we got notified that we made it to the Demo Day finals and took third place. Our COO, Nurai, is representing the project there.
Right now, I’m in the US taking part in the Hero Training program by Astana Hub and Draper University in Silicon Valley. What makes it different from other programs is that the focus is on the founders themselves, not just the product. Entrepreneurs are often put into unusual situations so the organizers can see how they handle uncertainty and pressure. For example, during one outdoor event, it was around 13°C, and famous venture investor Tim Draper suddenly jumped into a deep pool wearing a full suit. I jumped in after him and then pitched my product while standing in freezing cold water. We also recently came back from Survival Week, where we went through different challenges focused on resilience and leadership.
— Given your participation in the US program, are you planning to enter the American market?
— Definitely yes, but right now we’re still testing the waters. We’re heavily focused on Silicon Valley. There are so many entrepreneurs here, a huge startup and venture capital community, and a really strong culture of people helping each other. We’re planning to reach out to several speech therapy centers to understand their level of interest and how serious the problem is for them. Of course, we’ll need a different AI model for the US market, but ready-made models for English already exist. That said, the first version of the product will launch in Kazakhstan.