The present work explores the applicability of Variational Autoencoders in the search for a calibrator for implied volatility surfaces. To do this, two data sets provided by a financial entity were analyzed. In the first data set, Amazon, Euro Stoxx 50, S&P 500, Telefónica, and HKD were received. The best result was for the Standard and Poor’s (S&P 500) index model, with a mean absolute error on the surface of 1.29%. However, in the second dataset, the best result was for the combined model of all provided assets (Iberdrola, Inditex, Repsol, and SCH), reaching a mean absolute error of 1.21%. In addition, during development, encouraging results were found to explore possible use cases, such as synthetic data generation, scenario analysis, and cluster analysis.