Competition in the CompTech2021 week-long hackathon marathon. We created a tool based on thematic modeling that looked for similar articles/topics, documents, and analyzed the intensity of appearance articles over time. Unlike standard search engines, where you need to clearly and briefly write a query word for word, our model issues a query on the right topic, understanding that even if you write something approximate and the more words in the query - the more accurately it will give the necessary information. My task was to apply the BigARTM model to vectorize words, create a class that received these vectors and found n nearest neighbors using the NearestNeighbors model from sklearn. Despite the small composition of our team(2 people and curator Svyatoslav Kovalev). We have completed the task and protected the project. Project portfolio at the link.