The study of the structure of the proton has been an ongoing effort ever since its first experimental discovery in 1917 by Rutherford. As experimental methods progressed, and the properties of the proton, such as its mass, spin and charge, were uncovered, the question of its underlying structure arose. Experiments conducted in 1969 at SLAC undoubtedly confirmed the fact that the proton is not an elementary particle, but that it consists of smaller particles, which we now recognize as quarks and gluons. From that point on, physicists tried to understand how exactly the interplay of quarks and gluons inside the proton gives rise to its properties. One of the ways we can experimentally study the proton is through hard exclusive processes, such as deeply virtual Compton scattering (DVCS) and deeply virtual meson production (DVMP), which give access to a class of functions called generalized parton distributions (GPDs). GPDs describe the three dimensional structure of the proton, and they are a potential solution to the proton spin puzzle. It is currently not possible to extract GPDs from measurements directly, since they are convoluted with hard scattering amplitudes inside cross sections, and deconvolution is in itself an ill-posed problem. These functions are also not possible to calculate from first principles due to the non-perturbative nature of QCD, so we rely on modeling and machine learning methods in order to extract them from measurements. In this work we present recent results on the extraction of GPDs and relevant form factors from DVCS and DVMP data. Using the more recent proton DVCS data, we extract the set of leading Compton form factors (CFFs) with uncertainties and, by adding neutron DVCS data, we separate the contributions of up and down quarks to the CFF H. We make simultaneous fits to high energy deep inelastic scattering, DVCS and DVMP data at low x and study the impact of next-to-leading order (NLO) corrections, as well as the possibility of a unique description of these processes with twist-2 GPDs. We demonstrate a sizeable impact of NLO corrections and that extraction of unique GPDs becomes possible at NLO.