Yandex Neurosupport

The main articles are:

• Neural networks (neural networks)
• Data mining Data mining
• Speech Recognition (Technology, Market)
• Speech technology: On the path from recognition to understanding

2025: Launch of Smart Tip Generation Service for Contact Center Operators

On May 14, 2025, Yandex B2B Tech announced the launch of a service for generating smart prompts for contact center operators - Yandex Neurosupport. It performs the functions of a co-pilot: the neural network analyzes the text questions of customers and offers the operator an answer option. The specialist can choose: send the message unchanged, adjust the answer or compose it yourself. The service will help reduce the burden on contact center employees, as well as improve the quality and speed of service. The first Yandex Cloud customers are already using the service, in May 2025 it became available on request in closed testing mode.

As reported, Yandex has already introduced technology to improve support in its services. Thanks to this, "Food" and "Market" accelerated the solution of customer issues by 10-15%, and the level of customer satisfaction in "Food" increased by 15%. Among the clients of Yandex Cloud who are already piloting the service is one of the telecom operators, a housing and communal services company and others.

𝓲

Фото: Yandex B2B Tech

Yandex Neurosupport can be integrated into the API support service, and the service can also be deployed on the customer's own infrastructure.

We see the future of our ML platform in the development of application services that can be used without much code writing skills. For the most common business scenarios, we create end services such as Yandex Neurosupport or Yandex SpeechSense. For more specific tasks, a business can use our tools to create something of its own - for example, a highly specialized smart assistant based on the AI Assistant API. told Arthur Samigullin, head of the product ML-direction Yandex Cloud

The service analyzes information about the dialogue, the context of the situation and the knowledge base of the company to offer the operator the best answer. The prompt appears in the operator's work window for 1-2 seconds. He can immediately send it to the user, make adjustments or draw up his answer. According to the results of internal testing, operators use hints in every second dialog and in most cases make only minimal edits.

To generate hints, you need to upload the company's knowledge base to the service: this can be technical documentation, answers to frequent questions, instructions, CRM CRM systems data and training materials. During the dialogue, the neural network analyzes the client's question and previous messages, which allows you to comprehensively understand its problem and solve the problem.

Unlike chatbots, where artificial intelligence itself answers simple requests, a person who is helped by AI continues to communicate with the client in this service. So complex and complex customer issues are solved much faster and better. The neural network optimizes the work of operators and allows them to switch to more complex tasks. told Elvira Morozova, Head of Business Process Optimization at YandexGPT

In the future, AI agents will appear in the service - assistants who autonomously solve problems and interact with other applications. For example, they will independently collect and analyze information about the status of orders. Thanks to the introduction of agents, contact center operators will be able to receive more complete and up-to-date information without having to search it manually in various systems.

𝓲

Фото: Yandex B2B Tech

Yandex Neurosupport is based on lightweight models of the YandexGPT family, further trained on instructions for operators of more than 50 Yandex services. Unlike basic models, the service can effectively solve specialized support tasks, while providing a faster speed for generating prompts.

The use of Retrieval-Augmented Generation (RAG) technology allows the model not only to generate responses, but also to access internal data sources to find the necessary information. This provides a deeper understanding of the context and improves the accuracy of responses.

For example, algorithms automatically create a RAG index based on the knowledge base: they scan texts, identify keywords and phrases, and then create a "map" by which the system can instantly find the necessary information. When a request arrives, the system does not waste time sorting through all documents - it immediately accesses the index and finds the necessary information. Companies can quickly customize algorithms for specific documents and the flow of requests from their customers, which will allow the system to learn based on real requests and improve the quality of search over time.