Photo: Антон Леонардович Варфоломеев / Pexels
Igor Martynyuk, a 17-year-old entrepreneur from Kazakhstan, said he had raised more than $300 thousand for his startup TabAI. After moving to San Francisco, he shifted his focus to Heka, a platform that collects data to train artificial intelligence (digitalbusiness.kz). The new project aims to help AI work independently in computer applications.
What happened to Igor Martynyuk’s project
Martynyuk grew up on the outskirts of Pavlodar and learned to code using free online resources. In the summer of 2025, he joined nFactorial Incubator in Almaty. There, he built the first version of TabAI in two weeks and took first place at Demo Day, a showcase for projects.
TabAI was designed to bring together work-related information from email, messaging apps and task management services. Users could share this information with AI assistants and spend less time switching between apps. According to its founder, the product had more than 20 thousand active users at its peak.
Martynyuk said US investment firm Afore Capital invested $250 thousand. Nurdaulet Bazylbekov provided another $30 thousand, and Mark Vasilevsky contributed $5 thousand. He also named Andreessen Horowitz Scout Fund and angel investors among the other backers.
Why the founder switched to Heka
In Martynyuk’s view, developing TabAI was becoming harder as major platforms began adding similar features to their own products. The service also required users to change the way they worked. TabAI remains available, but is no longer under active development.
While working on the AI assistant, the entrepreneur encountered another problem: models struggled with tasks in graphical interfaces — windows, buttons and menus. Conversations with AI researchers helped him choose a new direction: collecting examples of people working on computers.
What this means
Heka users install an app and continue working in their usual programs. According to the founder, the service records their actions and mistakes, turns them into training data and sells it to AI model developers. Participants are paid for the data they provide.
“They can delete any part of it if they wish,” Martynyuk said, explaining how users can control the recording.
The entrepreneur’s next step is to build training environments based on the collected data. These will allow AI to try different actions in programs and correct mistakes. Martynyuk plans to release TabAI’s source code so users can further develop the product themselves.
