The key to making generative AI work on the RP2350 is reducing both the model's computational and memory requirements.