VK Created Discovery AI Neural Search for Multi-Billion Content Base of Zen and VK Video

VK published a detailed breakdown of the Discovery AI neural search architecture on Habr — a system combining a large language model with a multi-billion item content database of group services. The technology is being gradually implemented in Zen, VKontakte, VK Video, and Media Projects Mail.

What Discovery AI Includes

Discovery AI is not a single model, but a set of interconnected AI components developed by VK's Department of Experimental AI Technologies. Lead developer Evgeny Astafirov describes the system as a complex of five technologies:

• Neural search — search by query meaning, not keywords
• Context analysis — understanding the user's interaction history with content
• Personalization — adapting results to a specific person
• Generative models — synthesizing text responses based on found materials
• Recommendation algorithms — suggesting new content the user hasn't seen

The system is built on LLM integration with search indices and inference infrastructure. The system works with a database covering billions of content items from VK's entire ecosystem.

How Neural Search Differs from Regular Search

A traditional search engine finds exact word matches. Neural search understands intent: the query "suggest something about investments for beginners" returns relevant materials, even if those exact words aren't in them.

Technically this works through vector representations: the LLM translates the query and documents into numerical vectors, and the system finds content closest in meaning. This allows scaling search across billions of documents without checking each one.

The generative component adds the next level: instead of a list of links, the user can get a synthesized answer assembled from multiple sources within VK's database.

"Technology becomes an important component in developing recommendation and search systems in our products," —

Evgeny Astafirov, lead developer at VK.

Which Services and Who It's For

Priority platforms for implementation are Zen, VKontakte, VK Video, and Media Projects Mail. Beyond end users, VK plans to provide Discovery AI tools to content creators and businesses.

Gradual implementation means there's no single launch day: the technology will appear in different group services incrementally. No specific dates were mentioned in the publication.

What This Means

VK is building unified AI search infrastructure on top of its own multi-billion item content database — and this is a key competitive advantage unavailable to Western models. While Yandex develops similar systems, VK bets that deep context understanding within a closed ecosystem matters more than universality.

Frequently Asked Questions

Which VK services will neural search appear in first?

According to the Habr publication, priority platforms are Zen, VKontakte, VK Video, and Media Projects Mail. Evgeny Astafirov did not name specific dates for mass launch.

Will creators and businesses get Discovery AI access?

Yes — VK stated that Discovery AI tools will be available not only to end users, but also to content creators and business clients of group services.