VK builds Discovery AI, a neural search engine for the multibillion-item content base of Dzen and VK Video

VK has published a technical breakdown of Discovery AI, a neural search system that pairs a large language model with the multibillion-item content base of Dzen, VKontakte, VK Video and Mail. It combines an LLM, vector search, personalization and generative answers. The technology is being rolled out in stages across all of the group's services, for users, creators and businesses.

AI-processed from Habr AI; edited by Hamidun News
VK builds Discovery AI, a neural search engine for the multibillion-item content base of Dzen and VK Video
Source: Habr AI. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceHabr AI ↗
Editor-in-Chief
Neural searchsearch by the meaning of a query, not by keywords
Context analysisunderstanding the user's history of interaction with content
Personalizationtailoring results to the individual

VK has published a detailed breakdown on Habr of the architecture of Discovery AI, a neural search system that combines a large language model with the multibillion-item content base of the group's services. The technology is being rolled out in stages to Dzen, VKontakte, VK Video and Mail Media Projects.

What Discovery AI includes

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

  • Neural search: search by the meaning of a query, not by keywords
  • Context analysis: understanding the user's history of interaction with content
  • Personalization: tailoring results to the individual
  • Generative models: synthesizing text answers from the materials found
  • Recommendation algorithms: surfacing new content the user has not yet seen

At its core is the integration of an LLM with a search index and inference infrastructure. The system works with a base covering billions of content items from across the VK ecosystem.

How neural search differs from conventional search

A classic search engine looks for exact word matches. Neural search understands intent: the query "recommend something about investing for beginners" will return relevant materials even if they do not contain those exact words.

Technically, this works through vector representations: the LLM converts the query and documents into numerical vectors, and the system finds the content closest in meaning. This makes it possible to scale search across billions of documents without checking each one.

The generative part adds another layer: instead of a list of links, the user can get a synthesized answer assembled from several sources within the VK base.

"The technology is becoming an important component in the development of recommendation and search systems in our products," said

Evgeny Astafurov, lead developer at VK.

Which services and users it targets

The priority platforms for rollout are Dzen, VKontakte, VK Video and Mail Media Projects. Besides end users, VK plans to offer Discovery AI-based tools to content creators and businesses.

Staged rollout means there is no single launch day: the technology will appear in the group's various services gradually. The publication gives no exact dates.

What this means

VK is building a unified AI search infrastructure on top of its own multibillion-item content base, which is a key competitive advantage that Western models cannot match. While Yandex develops similar systems, VK is betting that deep understanding of context within a closed ecosystem matters more than universality.

Frequently asked questions

In which VK services will neural search appear first?

According to the Habr publication, the priority platforms are Dzen, VKontakte, VK Video and Mail Media Projects. Evgeny Astafurov did not give specific dates for a full-scale launch.

Will creators and businesses get access to Discovery AI?

Yes. VK said that Discovery AI-based tools will be available not only to end users but also to content creators and to business clients of the group's services.

Key Questions Answered

In which VK services will neural search appear first?

According to the Habr publication, the priority platforms are Dzen, VKontakte, VK Video and Mail Media Projects. Evgeny Astafurov did not give specific dates for a full-scale launch.

Will creators and businesses get access to Discovery AI?

Yes. VK said that Discovery AI-based tools will be available not only to end users but also to content creators and to business clients of the group's services.

What Discovery AI includes?

Discovery AI is not a single model but a set of interconnected AI components built by VK's AI Experimental Technologies Department. Lead developer Evgeny Astafurov describes the system as a combination of five technologies: Neural search: search by the meaning of a query, not by keywords Context analysis: understanding the user's history of interaction with content Personalization: tailoring results to the individual Generative models: synthesizing text answers from the materials found Recommendation algorithms: surfacing new content the user has not yet seen At its core is

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…