Machine-learned interatomic potentials (MLIPs) are transforming molecular simulations by providing highly accurate descriptions of molecular interactions, derived from quantum calculations, at a fraction of the computational cost (10.1016/j.cossms.2025.101214). While this technology has already revolutionized materials science and shows great promise for biomolecular studies, significant challenges remain in modeling chemical reactions. In particular, constructing chemically diverse and reliable datasets for a given reaction, or family of reactions, remains difficult. The internship will focus on optimizing data generation for the active learning of MLIPs, particularly for reactive events, by leveraging generative models such as Normalizing Flows (NFs) or score-based approaches, which can efficiently produce diverse molecular configurations. Unlike Molecular Dynamics-based sampling, generative approaches promise much greater computational efficiency. The candidate will gain strong experience in generative models and molecular dynamics simulations.