The simulations for in-ice radio detectors rely on analytically solvable exponential ice models, allowing for computationally fast ray tracing. However, these models do not encompass the whole complexity of the ice which may result in systematic errors on the simulations and reconstructions. We incorporated a numerical ray tracer module in NuRadioMC based on RadioPropa with accept-able speed and accuracy that focuses on accommodating more complex ice models. Presented here are the ray tracer’s implementation, and a first simulation and reconstruction using this new module. This shows that changing the ice model from an exponential profile to a double-exponential profile (firn model) has non-trivial effects on the timing. As this is important for distance and direction reconstruction of neutrino events, we hope to investigate this influence further once other effects like in-ice reflections or birefringence are implemented in this ray tracer as well.