Atom-diatom collisions play an essential role in cold molecular sciences. For instance, they are relevant to studying exotic Rydberg trimers or understanding the formation of molecules in a buffer gas cell. In this talk, I will present our research on long-range van der Waals trimers, a novel Rydberg molecule formed by a Rydberg atom attached to a polar diatomic molecule. We will explain the experimental conditions for its formation and describe its binding mechanism based on the van der Waals interaction. In addition, we will present our efforts to develop a universal chemical predictor.  We will focus on the Ca + H2 -> CaH + H reaction, an essential reaction toward forming cold CaH via buffer gas cooling. Our results show that a machine can efficiently learn the product state distribution for reactive and inelastic processes. Furthermore, and more impressive is our finding that our machine learning model learns even across the isotopologues space in chemical reactions.