Yandex has introduced the Neuroexpert service, which is available in beta version. It allows you to upload documents, tables, presentations, audio and video files, as well as send links that form a knowledge base with the ability to find an answer to any question. Anyone can use the service.
Image source: Yandex
The knowledge base on the NeuroExpert platform is similar to a folder in cloud storage — it can be shared via a link. The user uploads information and asks questions about it in dialogue mode; the system prepares answers only based on the provided data. During the beta testing stage, up to 25 files and links are uploaded — these can be documents, tables, audio messages from a messenger, or other data. Once the beta testing is complete, more source data can be added.
The service will be useful for those who work with large amounts of information. A teacher can upload their lectures and pass the database to students, who will find answers to their questions in it when preparing for exams. NeuroExpert will be useful, for example, if you need to create a report, and the necessary data is located in one of the many presentations – you will not have to search for information manually. The service will help you make a purchase decision if you upload links with product descriptions and video reviews. Finally, you can upload board game rules to NeuroExpert and ask him a question if a controversial issue arises.
The service will be especially useful for business clients – it can be connected to internal knowledge bases and documentation. “NeuroExpert” will help systematize work, will facilitate adaptation and increase the efficiency of the company’s employees. You can connect a preliminary version of the corporate “NeuroExpert” upon request.
The service uses several technologies created by Yandex. The visual language model (VLM) is used to search for data in graphs and diagrams; speech recognition technology (ASR) is responsible for processing audio and video; optical character recognition technology (OCR) processes text in images. The large language model YandexGPT 5 Pro processes data from these systems and prepares a response: knowledge of the world order and language rules helps it understand the request and prepare a high-quality response – while it takes the actual data from the uploaded materials. Another component unites the knowledge of the model and information from the files – the RAG system (Retrieval Augmented Generation).
Foreign developers already have analogues of NeuroExpert: Google NotebookLM, Perplexity Spaces and ChatGPT Projects. Yandex plans to implement two models of service monetization: extended capabilities for users who have subscribed and integration of NeuroExpert into the customer’s information systems.