This proposal establishes an industrially relevant methodology for operando characterization of homogeneous and heterogeneous reactions under harsh conditions in gas and liquid phases currently unavailable for regular users. The scientific cases will be based on two classes of novel catalytic systems: Ru-mediated defunctionalization of polyols to olefins and alkenylation of arenes via direct C-H activation over single-site Pd-catalysts. Initially, a spectral database of well-defined Pd and Ru compounds will be collected and used as a training set for machine learning. Then, we will step by step increase the complexity of experimental conditions from currently available cells to a reactor that can withstand up to 250°C and 50 bar, with the possibility to sample the gas phase and carefully dose liquid reactants. Finally, the ML-based system will be implemented and tested allowing for online evaluation of structural and catalytic data and automated refining of the reaction conditions.