Dynamics and solvation effects are fundamental in modelling chemical processes in liquid phases, including homogeneous catalysis and biochemical reactions. The reaction environment critically influences the structure and stability of participating species, thereby determining reaction rates, selectivity, and mechanistic pathways. Despite their importance, accurate computational modelling of these effects remains challenging, particularly for polar solvents, where explicit solute-solvent interactions must be captured at a high level of theory, such as hybrid DFT and beyond.

In this talk, I will present our development of reactive machine learning interatomic potentials (MLIPs) designed specifically for modeling chemical processes in solution. Our methodology integrates automated active learning with enhanced sampling techniques and descriptor-based structure selection to create data-efficient training sets that accurately reproduce the DFT reference. By combining the Atomic Cluster Expansion framework with either linear regression or message-passing neural networks (MACE), we demonstrate how MLIPs significantly accelerate molecular dynamics simulations of solution-phase reactions, enabling the modelling of chemical processes under experimentally relevant conditions.