The beauty of mpy is that it is bare metal. It would therefore be nice to explore this in four directions:
1) a real low power version, able to use a harvester, minimum software, hardware, charger, etc.
2) a version running on a fpga platform e.g. Zynq...and use python to generate some vhdl...
3) a dsp version
4) a multiprocessor version (how about stacking a few pyb boards

1 and 2 would be great. 3 and 4 can be created from 2...
Regards, Roland