Determination of stereochemical configuration in organic compounds may be very often helped by the prediction of chemical shifts or scalar couplings through DFT quantum-mechanical methodologies. However, these computations may easily become a bottleneck on the structural elucidation process. We present here different DFT-mimicking machine learning approaches, using either kernel-ridge-regression or neural-network methodologies, applied to the prediction of the more relevant scalar couplings in structural elucidation namely geminal 2JHH and 2JHC as well as vicinal 3JHH and 3JHC ones. A large dataset of B3LYP/6-31G* scalar couplings ,computed on a miscellaneous collection of molecules retrieved from the PubChem database, was employed for training and testing while molecular representations consisted of purely geometry parameters combined with electronic parameters. The performance of the here developed algorithms improved that of known empirical equations while presenting a much broader degree of applicability. Pathological couplings in hydrocarbons found during exploration of results will be discussed. Delta approaches based on cheap DFT computations for the prediction of 13C isotropic shieldings and chemical shielding anisotropies will be also briefly presented.