The paper considers the task of developing question-answering systems that provide consultation to users based on the data from large amounts of text documents that weren't previously used for training the developed models. The paper provides the results of the application of the retrieval-augmented generation method and large language models in question-answering task according to text documents. To solve the task, the initial information is divided into fragments. Each fragment is assigned an embedding, which is used to find the proximity of this fragment to the user's question. The results of applying embedding models, choosing a vector database, and the architecture of the developing software are also presented in the paper. Based on the results of the study, the authors developed the software prototype of question-answering systems, providing consultation to clients of a travel company on issues of choosing tours.