The implementation of nonlinear, transient multiphysics problems often suffers from long development cycles due to the manual derivation of residuals and Jacobians. In this talk, a new high-level interface for the open-source library AutoPDEx [1] is presented, which allows a rapid prototyping of multiphysical models without significant implementation effort. The framework is built on JAX [2], a code-to-code transformation library that enables automatic differentiation and just-in-time compilation, including support for GPU hardware. By leveraging these features, AutoPDEx allows users to define the physics of a problem directly through its weak form as a JAX-traceable function. For spatial discretization, finite-element-based approaches are employed. For temporal discretization, the user can choose, e.g., from Newmark, Adams-Moulton or Backward Differentiation Formula schemes. The functionality of the interface is demonstrated through academic benchmarks and geomechanical applications.