Optimizing the in vitro transcription (IVT) process for mRNA production of large mRNA molecules is a challenge. To address this, an active learning approach was adopted, where first the experimental data is analyzed and the process modelled and then the derived insight used to support designing the next iteration of experiments. A hybrid modeling strategy was chosen, allowing to use the available fundamental knowledge while accounting for changes in the process that would be difficult to incorporate into a more mechanistic approach. Also, novel experimental evidence or changes to the process can flexibly be integrated into the model. In total, five experimental iterations were performed with the objective to explore the process conditions in which integrity is above a specified threshold while also considering yield and process costs. The hybrid models showed excellent performance across the iterations, allowing to derive detailed understanding at every iteration, e.g. using what-if simulations or Shapley value analysis, which are presented in a qualitative manner. The model structure is such that additional inputs could be added in subsequent activities for scale-up support or the development of processes for other mRNA molecules.