My Vibe & AI-comments

Prerequisites for reimagining the music experience

• ✦ users passively listen to music (play a track and put their phone in their pocket) without engaging with the service's scenarios
• ✦ listeners don't discover new music, don't seek out anything new, which led to feedback that the recommendations often suggest repetitive tracks
• ✦ one of the business goals is to help up-and-coming musicians grow, meaning people need to discover their music

Mission

Enable discovering new music and musicians, feeling that the app understands you and your context, recommending exactly what you need right now

Goals

• ✦ growth and retention of LTV
• ✦ trust in recommendations and increase in listening session per user
• ✦ strengthening the brand
• ✦ growth in loyalty
• ✦ increase in retention
• ✦ artist-centric approach and growth in concert ticket sales

What was achieved

In May 2026, the new My Vibe was launched. The music recommendation system is powered by the generative model ARGUS (Yandex presents a method for training large transformer recommenders with up to 1B parameters). The model analyses a longer history of user actions, finds non-obvious connections between actions, and takes external factors into account. ARGUS analyses not only listener preferences, but also external context: time of day, day of the week and which device the music is playing from. For example, if a listener loves jazz, opens Yandex Music on a laptop, and it's a weekday, My Vibe will suggest «calm jazz for working».

Launch results

• ✦ likes in My Vibe increased by 4%
• ✦ dislikes decreased by 2%
• ✦ click-throughs to artist pages grew by 18%

More about AI-comments

I was also responsible for integrating AI into My Vibe. The goal was to create not just another chatbot, but to find a way for the LLM model to help people engage more deeply with artist's work without disrupting the music experience.

Goal

To immerse yourself in an artist's work, learn more about the performer and their music, uncover the meaning of a track or the artist's intent, and connect it to yourself.

Work Process

• ✦ during the discovery phase, conducted interviews with different listener segments
• ✦ based on the data, vibe-coded a high-level concept in cursor and xcode
• ✦ presented the concept to stakeholders, gathered feedback and open questions

What was done

Alongside the new My Vibe, we launched AI-comments that share interesting facts about the performer and their music, reveal the meaning of tracks, and draw listeners into the artist's work.

• ✦ we display an AI-comment on the My Vibe screen, condensed to 100 characters
• ✦ we collect feedback

Variety of facts and personalisation

AI-comments can be not only about the artist and their work, but also about the user themselves, in order to create a stronger emotional connection with the product. For example, the LLM model knows that the user has tickets to Istanbul booked for next month and proactively suggests attending a Kanye West concert on those dates.

AI-comments launch results

• ✦ concert GMV increased by 4%
• ✦ conversion to artist donations increased by 4%
• ✦ 16% of users open AI-comments to read the full comment
• ✦ in feedback on AI-comments, likes outnumber dislikes by 3 to 1

Future plans

We continue to receive feedback, both positive and negative, analyse it, and make improvements. Alongside this, we are working on new development tracks for the new My Vibe, which we plan to launch in the near future.

My role

• ✦ concept development
• ✦ AI-comments from scratch
• ✦ working on My Vibe on web
• ✦ working with the My Vibe player
• ✦ developing micro-animations for interactions