Type II supernovae provide an independent way to probe cosmology. Due to the comparatively simple physics of their hydrogen-rich envelopes, distances to these objects can be inferred directly from first-principle radiative transfer modeling. Up to now, the high computational costs of radiative transfer calculations, as well as the need to fit the data through optimization by hand and eye have prevented the wide-spread application to cosmology. To tackle this issue, we have used machine learning techniques to develop a spectral emulator to replace the simulator i.e. the radiative transfer code during the fit process. We train the spectral emulator to predict the output of our radiative transfer code TARDIS to high precision, based on a set of pre-computed spectra. The trained emulator is orders of magnitude faster (~1 ms) than TARDIS (~100000 s) -- this makes it possible to fit spectra automatically. We present results for modeling spectra of SN1999em and SN2005cs using the newly developed approach.