In evolution, proteins adapt and diversify their sequences through a complex interplay between random mutations and selective pressures. This process results in protein sequences with a shared evolutionary origin that retain nearly identical three-dimensional structures and functions, despite differing by as much as 70-80% in amino acid composition. In this presentation, I will discuss our work on modeling protein evolution across multiple timescales, spanning from the impact of individual mutations to patterns observable over deep evolutionary epochs. Here, I will present our approach to modeling this evolutionary process across multiple timescales, from single-mutation events to changes across deep evolutionary time. We leverage methods rooted in inverse statistical physics to construct fitness landscapes, which capture the mutational impact on protein functionality and even allow the generation of synthetic yet functional protein sequences. Furthermore, we model evolution as a stochastic process traversing these landscapes, emphasizing the collective dynamics in protein evolution. This collectivity manifests via an emergent separation of timescales: a fast mutational timescale within a static sequence context and a slow timescale associated with the gradual shift of the sequence context itself.