The introduction of electronic medical records of patients, the development of a unified state information system made it possible to accumulate large amounts of data that can be used for the implementation of artificial intelligence systems. This approach will reduce the time for preparing documentation and helps increase satisfaction with the quality of medical care. Within the framework of this study, an attempt was made to consider an approach to the analysis of unstructured Russian-language medical texts as part of the solution to the problem of generating recommendations for the patient, based on the GPT deep learning model, previously trained in a Russian-language text from sberbank-ai. The loss value on the validation data set was 11.16. The assessment of the similarity of the generated recommendations with the real recommendations of doctors based on the BLEU metric based on bigrams and trigrams averaged approximately 0.333 and 0.107, respectively. The results obtained demonstrate the potentially high possibilities of using NLP methods to build medical decision support systems.