Kazakhstanis Invested 120 Million Tenge in an AI Service for Teachers Already Used by 13,000 Educators
Madiyar Yergali and Yelaman Bauyrzhanuly have spent the past two years building Shyraq AI, a startup that helps teachers put together lesson plans, games, videos, presentations, and other interactive materials for their classes.
As part of Digital Business and Astana Hub’s joint project, “100 Startup Stories from Central Eurasia,” Madiyar talked about why the team is not looking for investors, whether it is ethical to make money from a tool for teachers, and how they deal with AI hallucinations.
“Teachers pour their time and energy into helping other people’s kids grow, but they still don’t get the respect they deserve”
— Tell us a bit about yourself. What were you doing before Shyraq AI?
— I’m from Taraz and studied at a Kazakh-Turkish lyceum. That’s where I first got into building new things: I took part in computer science and robotics competitions, and came third at Infomatrix, an international robotics project contest. That was when I realized how exciting it can be to solve tough problems with IT.
Then I got into Nazarbayev University to study Computer Science, and by my third year I had started working in web development and building Telegram bots. After graduating, I was immediately offered a remote job at a company in Armenia. I worked there as a developer for six months and realized I didn’t like working for someone else. So a classmate and I opened an education center in Taraz, kind of an after-school program where students could improve their math and English. We invested in renovating the place, but unfortunately we couldn’t attract enough clients. We had to close the center, and I spent a long time paying off the debts afterward.
In 2023, when the AI boom kicked off, I decided to try my hand in that space and got a job as a data scientist at a bank. I picked things up quickly, moved up a level, and realized that a lot of routine tasks were pretty easy to automate. In the end, I was only spending two or three hours a week on actual work, and used the rest of the time to keep learning.

Madiyar Yergali
— Did the bank know you were working that way?
— No, but it wasn’t really a problem. A bank is a huge machine, and it keeps running steadily even if some individual employees aren’t super involved. A lot of specialists just handle the routine tasks in their area, but I didn’t want to spend my time doing that. So I automated my work, while still getting everything done: analytics, data collection, and data engineering tasks.
— Why did you leave the bank?
— Active people tend to get bored pretty quickly in banking, because in a big organization, every change takes forever to get approved. I wanted to grow faster and build something new.
I left to teach at a private school, and in my free time I took on outsourced projects, helping companies add AI features to their apps through APIs. Later, I spent three or four months in a solid role at FUN&SUN, where I worked on an internal AI-powered document management system and a corporate RAG system. RAG stands for Retrieval-Augmented Generation, a technology that connects AI models to a company’s documents or data.
Even though I only spent one semester working at the school, it made me realize that teaching is just as demanding and carries just as much responsibility as being a doctor. In Kazakhstan, teachers still don’t get the recognition they deserve, even though they pour their time and energy into helping other people’s children grow. That’s when the idea came up: why not build a service that helps solve teachers’ problems?
Around that time, I met my current partner, Yelaman Bauyrzhanuly, and we started building Shyraq AI together.

Yelaman Bauyrzhanuly
— How did you choose your business partner?
— A mutual friend introduced us. He told me he knew a great sales and marketing person who wanted to launch an IT project, and suggested we team up. We had a chat, and now we’ve been working together successfully for almost two years.We split the responsibilities right from the start: I’m in charge of the product and tech, while Yelaman handles sales and marketing.
My advice to any IT specialist who wants to move into business is to find a partner who balances out your weak spots. Developers tend to get so deep into the technical side that they don’t have much energy left for sales. And for that, you need someone energetic, someone who’s good with people and knows how to explain the value of the product. That’s exactly what happened with us. Plus, Yelaman had already run a successful business for teachers, so he understood this audience really well.
“Cut the time teachers spend planning lessons by 90%”
— What problem does Shyraq AI solve for teachers?
— First and foremost, it helps with lesson planning and paperwork: things like short-term lesson plans, original teaching programs, and research projects. We used AI to automate those processes, cutting the time teachers used to spend on planning by around 90%.
— Does a teacher need to know how to write prompts for that?
– No, and that’s a big part of the idea. We deliberately moved away from the “plain ChatGPT” approach, where you need to know how to phrase a prompt properly. In Shyraq AI, a teacher just selects the subject, grade, language, and lesson topic from drop-down menus, then hits “Generate.” Three to five minutes later, they get a complete short-term lesson plan in the exact official format required by the Ministry of Education. It includes the learning objectives from the curriculum, the lesson objective, expected outcomes by level, all five stages of the lesson, assessment criteria with descriptors, differentiation, cross-curricular links, and values. Then they can export it straight to Word, print it out, and hand it in to the head teacher without changing a thing.
– How do you make sure the system isn’t just pulling random text from the internet?
– We don’t use fixed templates from a database, and we don’t rely on “raw” AI that just makes things up. Our system is a hybrid. Shyraq AI has Kazakhstan’s official textbooks uploaded and indexed in advance, broken down by subject, grade, and language. So when a teacher generates a lesson plan, the system uses RAG technology to pull real tasks and materials from the official textbook, with page numbers included, and it can even bring in images of the exercises.
That’s where the roughly 90% time saving comes from. Before, a teacher might spend an hour and a half or two hours putting together a lesson plan. Now it takes about five minutes, and the plan is based on a real textbook, not generic filler from the internet. Of course, the teacher’s own input still matters. They should read through the plan and make any edits if needed.
For teachers who have no time at all, we offer ready-made sets of short-term lesson plans, end-of-unit assessments, and end-of-quarter assessments. They were put together by professional methodologists and curators with master’s degrees. And if a teacher needs something beyond the ready-made materials, they can use the AI tools inside the platform to generate presentations, documents, and games themselves. If they have trouble using a computer or smartphone, our curators hop on a video call and show them how everything works.
— What other tools does Shyraq AI offer besides lesson plans?
– We’re adding more interactive features, like AI-powered educational games and activities teachers can use in class. Here’s how it works: the teacher just enters the lesson title and topic, selects the goals and textbook, and chooses the game they want the AI to create for that specific lesson. There’s also a prompt field in case they want to customize anything. Then they just hit the button, and everything is ready in three to five minutes.
We also have simulations for lab work and experiments in physics and chemistry. Now we’re building similar tools for biology, geography, and Kazakhstan history. This is especially important for schools that don’t have the resources or equipment to do real experiments.
We also have Ybyrai Bot, which is optimized for Kazakhstan’s education standards and helps teachers quickly find accurate answers, Túlğa, a service where students can talk to famous historical figures, and a presentation builder for lessons.
One really important feature is that teachers can download all the materials and use them offline. That’s especially helpful for teachers in rural schools, where internet access can be unreliable.
— How do you deal with AI hallucinations?
— No one in the world has fully solved that problem yet. From what we’ve seen, AI is most likely to make things up in humanities subjects, especially when it’s dealing with older literary works whose texts aren’t available online. In those cases, we recommend that teachers take photos of the relevant pages from the book and attach them to the request as context.
Our methodologists also review the generated content from time to time, and we adjust the prompts when needed. The RAG system helps too, because it lets the AI pull information strictly from approved school textbooks.
But the most important thing is that teachers personally review the content before showing it to students.
— Tell us about your team and the tech stack you use.
—.Our IT department is made up of three people, including me as CTO, but we do the kind of workload you’d normally expect from a team of ten. We also have around ten curator-methodologists, five marketers, and 20 sales managers. At our peak, when we pooled resources with my partner’s other business, we had 70 people on the team, including 13 in IT together with the interns.
On the tech side, we use a mix of models from OpenAI, Google, and Anthropic. We tried training our own local LLMs on GPU servers, but for mass use, it just doesn’t make financial sense. Maintaining our own servers would require a huge investment, and it wouldn’t pay off when working with teachers. So now we combine different models depending on the task: some are faster, others are cheaper. We also use third-party services to generate character videos for the Túlğa module.
“Making big money off teachers feels morally questionable”
— How does the project make money?
— We use a subscription model, either monthly or yearly. The price depends on the plan: there’s a standard limited subscription, and there’s also a credit-based option where users aren’t tied to strict feature limits but spend internal credits instead. For example, someone could use all their credits to generate 200 short-term lesson plans at once.
We’ve now switched to the credit system and cut the price of the basic plan in half, from 10,000 to 5,000 tenge a month. We’ll see what demand looks like in the new school year.
Honestly, if teachers used their limits to the fullest, we’d be running at a serious loss. But most of them use the platform only when they actually need it, which helps the unit economics balance out. Plus, usage peaks at the start of the school year and then drops off, so overall we still come out ahead.
— Are you actually making money from it?
— We are making a profit, but it’s not huge. For about a year and a half, we were operating at a loss. Even now, during the low season, we sometimes dip into the red, but on average over the year, we break even or make a small profit. Still, the project has real social value because it genuinely helps teachers save time.
In general, making big money off teachers feels morally questionable. That’s why we’re also building another startup in parallel, Xrox, a platform for automating communication on Instagram, WhatsApp, and Telegram through chatbots. We already have major clients with turnover of around one billion tenge who actively use the system to generate leads.
As for Shyraq AI, we’d like to make the platform as accessible as possible for teachers across Kazakhstan. That’s why we’re now trying to build relationships with the government.
— How would that work?
— We want the Ministry of Education to buy the system centrally for teachers, so they don’t have to pay for Shyraq AI out of their own pockets. For many teachers, even a few thousand tenge a month is a real expense. Some schools are already buying access to our platform through government tenders, at the request of regional education departments. That experience shows the model can work.
— How many teachers are using the platform?
— Around 13,000 use it regularly. Out of more than 400,000 teachers in the country, about 155,000 follow us on Instagram. We have 65,000 registered users in the system, and 58,500 of them have tried the demo version, so they already know what the platform can do.
And we feel like we’ve more or less reached the ceiling of the paying teacher audience in Kazakhstan. So the next logical step is to move to the government level.
“We tried training our own local LLMs, but it didn’t make financial sense”
— How hard was it to get teachers to pay for a subscription in the beginning?
— Yelaman had already been working with teachers for a year and a half before we met, helping them pass certification and write original teaching programs and professional development journals. So his experience gave us a strong start: we understood the audience’s pain points right away. But we still had to build the customer base from scratch, and the hardest part at the beginning was getting the first positive reviews and earning people’s trust.
— What customer acquisition channels work best for you right now?
— Instagram, TikTok, and word of mouth. Over the past six months, word of mouth has really taken off because teachers have been officially required to learn AI tools. So they’re actively looking for useful services, and when they find one, they recommend it to each other.
— What makes Shyraq AI different, given that teachers can just generate things in ChatGPT themselves? Especially now that the AI giants are quickly adding new features directly into their own products.
— Our main value isn’t just generating content in exchange for tokens. Shyraq AI’s real strengths are that it’s local, it comes with hands-on support, and it’s adapted to the teaching standards of a specific country. We help teachers do more than just use our platform. We help them grow, run blogs, create media, and keep up with technological progress.
Of course, the big AI models will keep evolving, and we’ll keep evolving with them by building their new capabilities into our product. I think companies that know how to hold on to their customers, provide great human support, and adapt to change can get through any crisis.
— How ready are teachers in Kazakhstan to use AI and create interactive materials for their lessons?
— Many teachers still want everything done for them. So our curators often have to explain that the platform is a tool for their own creativity, not a full replacement for the teacher.
But there are also some really proactive teachers who picked up the system quickly. They started running training sessions in their schools and regions, launched their own blogs, and now help other teachers put together short-term lesson plans and original teaching programs as a paid service.
We reward the most active ambassadors of our service by giving them honorary Shyraq AI partner badges.
“We’re hesitant to enter new markets because we don’t want to lose ground in Kazakhstan”
— How much have you invested in Shyraq AI? Have you raised any funding?
— We’ve been building the company with our own money. In the beginning, my partner put in around 60 million tenge from his other business. We also took out loans: about 20 million tenge last year and around 40 million tenge this year. We’ve paid all of that off now, and the original investment has already paid for itself.
— Do you need an investor now?
— No, we’re not looking for investors. But we’re open to partnerships if we expand into CIS markets through a franchise model, as long as we find partners who are ready to cover the operating and marketing costs in those regions.
— Why a franchise model specifically?
— For now, we’re hesitant to enter new markets on our own because we don’t want to weaken our position in Kazakhstan. There are plenty of examples of startups that overestimated their capacity during international expansion and ended up failing.
For us, Shyraq AI is a project that lets us contribute to Kazakhstan’s development. We make money through other IT ventures.
— Where do you see the project in the long run?
— We’re getting ready to expand to students, and we’ll be launching the first services at the start of the school year. Right now, we’re actively adapting some of the world’s best teaching practices to our education system, including Singaporean methods built around teamwork. So soon, Shyraq AI will offer high-quality interactive learning modules that children in Kazakhstan can use outside the classroom.
More broadly, we’d like to build a full education ecosystem: a super app that brings together teachers, students, and parents. It could even go as far as a marketplace for school supplies, which we could launch with partners. I think the next three years will be enough to see whether this model really works.