While continuum models such as the Navier-Stokes equations begin to break down as the Knudsen number (Kn) approaches unity [1], the Boltzmann equation (BE) provides a kinetic description of gases across the entire range of Knudsen numbers. It was originally derived from Molecular Dynamics by letting the number of particles approach infinity under the Boltzmann-Grad limit [2]. This derivation gives rise to a collision operator which is consistent with conservation laws and entropy dissipation by definition. This operator has an exact formulation up to an unknown collision kernel. Traditional formulations of this kernel have relied on their mathematical properties such as detailed balance and non-negativity. Using these properties, the collision operator can be proven to dissipate entropy [3]. However, these kernels often significantly simplify the actual molecular interactions. In the present work, we propose a data-driven approach to fit a parametrized collision kernel to high-fidelity Molecular Dynamics data. We thereby aim to exploit the fundamental relationship between the Boltzmann equation and the MD formulation from which it is originally derived. Conservation of mass, momentum and energy, as well as entropy dissipation are guaranteed a priori by the mathematical properties of the chosen parametrization. We facilitate the simulation of polyatomic gases by extending phase space with an additional dimension, representing the internal energy of the particle. The collision operator can be extended accordingly, using the Larsen-Borgnakke parametrization of collisions [3]. To solve the resulting Boltzmann equation, we employ the Discontinuous Galerkin Method of Moments (DGMoM) approach [4]. This computational framework combines the spatial discretization of the Discontinuous Galerkin Finite Element Method, with the velocity and internal energy discretization of the Method of Moments [5]. The macroscopic quantities of the gas - density, bulk velocity, temperature, etc. - appear as moments of the distribution function, which allows for comparison with experimental results.