Explore a conference talk from JuliaCon 2024 that delves into the evolution of WaterLily.jl, a computational fluid dynamics solver in Julia. Learn how this CFD solver transitioned from a serial-CPU implementation to a backend-agnostic solution capable of seamless execution across multi-threaded CPUs and various GPU vendors. Discover the meta-programming approach used to generalize array iterator implementation and the utilization of KernelAbstractions.jl for architecture-specific kernel specialization. Examine performance comparisons showing WaterLily.jl matching state-of-the-art CFD solvers written in C++ or Fortran in single-GPU tests. Gain insights into the potential integration of machine learning models and differentiability into the solver, expanding its capabilities for future applications.