Yandex AI is a new AI app for the Turkish market. It consists of three parts: the Chat itself, the Feed, and the Browser.
The AI Chat is the app's core functionality. The Feed is a place where we educate users on app features and display personalized news, products, and offers. It is used to boost retention and as another entry point into the AI Chat.
TLDR
Role and impact
I worked as a Lead Product Designer in a team of five Product Designers and a dozen PMs.
- I launched a generative news video system from scratch and helped scale it to hundreds of videos per week;
- Turned the Feed into a set of entry points to chat;
- Worked on the Chat response formats for News, Sports, and Earthquakes;
- Led the work for Onboarding project, created 10+ concepts, and conducted 8 UX studies. What a project;
Below, I describe all of this work in a bit more detail.
Generative News Videos
2025
250+
videos per week
Problem
At the start, this format did not exist. There was no pipeline, no clear quality bar, and no good way to turn mostly text news and weak images into videos that felt trustworthy.
Solution
I developed a design system for video production, came up with a product implementation idea, and created the first proof-of-concept prototype in code, using AI. After that, I formulated a JSON contract for data transfer and a Quality Tier system for internal quality assessments.
Result
The system went into steady production and started making hundreds of videos per week. It covered major news and many mid-size stories, and later other teams used it too, including Yandex Maps.
Feed
2025
9 → 43
content types in the feed
Problem
At first, the feed was mostly a layout. It was not clear what job it should do next to chat, and the design did not hold up well when new content areas were added.
Solution
The feed was meant to be a place of inspiration. Each card had to act as a point of interest and spark user curiosity. To achieve this, we carefully selected the most important ones and diversified the content mix.
I also developed a system and a set of rules for the cards. My experience at IVI and Yandex Games helped a lot.Result
The feed grew from a small set of individual cards into a system that could scale. We also expanded the range of content that engaged users and led them into chat.
News, Sport, Earthquakes in Chat
2025
Problem
The basic quality of responses to news, sports, and earthquake queries was low and uneven. Responses were difficult to scan, often vague and long or, conversely, too brief and direct.
Solution
We explored different types of user queries. Based on this, I created structured responses for them. I used flexible layouts built from our design system components and created new ones when needed.
Result
We've significantly improved the quality of our responses to people's key news queries, and based on research, we've caught up with our competitors, and in some places, even surpassed them.
Onboarding
2025
10+
concepts
8
studies
Problem
The team did not agree on what onboarding should do first. The launch was close, priorities kept shifting, and the product itself was not easy to explain in a few screens.
Solution
I worked through the direction from ideas to research. After more than ten concepts, we landed on a two-part flow: a short intro video, then a lightweight onboarding around interests and the first action.
Result
The final version shipped on time after eight qualitative studies. It gave the team a grounded solution for launch and left room for future versions, including chat-based onboarding.
Outro
These projects primarily involved creating systems, not just screens. I strived to maintain a high standard of quality despite the tight deadlines.
Ultimately, we launched the new app in eight months.
