In this paper, we discuss the development of a nuclear data evaluation pipeline, based around the TALYS code system. The pipeline focuses on the evaluation of the fast neutron energy range, above the resolved resonances. A strong focus in development lies on automation and reproducibility, as well as the efficient use of large-scale computational infrastructure, to enable rapid testing of new algorithms and modified assumptions. Several novel concepts for nuclear data evaluation methodology are implemented. A particular problem in evaluating the neutron-induced reaction cross-section using TALYS, relates to the intermediate energy range. While TALYS only predicts the smooth energy-averaged cross-section, experiments reveal unresolved resonance-like structures. In this paper, we explore ways to treat this type of model defect using heteroscedastic Gaussian processes to automatically determine the distribution of experimental data around an energy-averaged cross-section curve.