Schematic representation of OssicoNN, the neural network for the inference of stellar parameters. The primary objective of the neural network is to encapsulate a transformative mapping between the hidden stellar parameters and an abstract latent space composed of an equivalent number of Gaussian distributions. This is done by the conditional network (in blue). The mapping is conditioned by the spectra first processed by the conditioning network (in red) that extracts the important features from the spectra. Each layer is shown as a box in a different colour, reflecting its nature, with key hyperparameters listed beside it.