Can multiple Architectures be placed sequentially together?

I would like to use FNO/AFNO with mixed data with different sizes, i.e. time-series with matrix input for variations.

I am aware that these architectures only accept 1D/2D/3D structured (AFNO-2D) data but not mixed. To provide the 1D time-series data in 2D form seems to be a little excessive, thus I would like to use another architecture such as DeepONet/fully_connected,etc to encode this time-series data into a 2D latent space, before passing it into the FNO.

This however requires that the output of one architecture can be fed as input into another, and the whole system trained with a single validatior. The examples I currently see treat multiple architectures as independent mechanisms (or so I think)

Can anyone help and point me in the right direction.

Thanks in advance