The upcoming advent of exascale computing poses unique scalability challenges to scientific applications, as millions of cores and hundreds of thousands of GPUs will likely be available for dedicated at-scale calculations. While some of these calculations will involve monolithic codes, many other exciting applications will take the form of complex workflows orchestrating billions of interdependent tasks. The EXAALT project, funded by the US Department of Energy, is tasked with developing exascale-scalable atomistic approaches to autonomously develop kinetic models of materials. The fact that the corresponding workflows are composed of extremely large numbers of very short and tightly coupled tasks makes these applications particularly challenging. I will introduce our computational framework and demonstrate how its modular design can be leveraged to implement a number of complex algorithms, from long-time molecular dynamics, to reaction network construction, to data/machine-learning driven workflows.