A neural network developed in Russia, named GENATATOR, can "read" DNA and automatically build a gene map, which can speed up work with the genomes of organisms for which there was previously little biological data. The system examines a DNA sequence to determine where genes begin and end, what types they are, and how they are structured internally, a complex task because genes lack clear "markers." Unlike older methods using predefined rules, the model was trained on a large number of genomes, allowing it to find common protein-coding regions and more complex genes like long non-coding RNAs. The technology is especially useful for "non-model" organisms with only raw genomic sequences available. During testing, GENATATOR, trained on humans and 38 mammalian species, successfully analyzed data from other living organisms, including Drosophila, plants, and yeasts, and found rare "poisonous" exons that can destroy RNA. An open leaderboard was created to check the model's quality, where GENATATOR shows good results, and the model is available on the Hugging Face platform.