Abstract
The article presents the initial stages of developing an intelligent information retrieval system on the website of the FRC KSC RAS, based on the use of large language models, ontological and prompt-engineering, and RAG-technology. The system is aimed at increasing the relevance of responses to user queries formulated in the dialogue process in natural language and related to the website content, as well as to the electronic repository of documents of the FRC KSC RAS. The system architecture has been designed, and a prototype has been developed that implements some of the proposed features of intelligent retrieval. The use of RAG-technology allows for analysis of current context without the need to retrain a large language model, which is especially important in situations where information on a website can be added and modified ad hoc. For some query types, specialized prompts have been developed to allow you to formulate an answer that corresponds to the specific of the query. In order to prevent the transfer of confidential information to external services, it was decided to deploy the software system locally on the server of the IIMM KSC RAS. The experiments carried out demonstrated the feasibility of the proposed solutions for the considered tasks of intelligent retrieval.