That's quite a techy situation you're wrestling with! Building with TFLM can sometimes feel like herding cats. I can see you're diving deep into embedded ML which is cool. Ever think about optimizing kernels manually for performance? I remember a time when I was trying to get a neural network to classify different types of fruit on a Raspberry Pi, resource management became critical, and everything slowed to a crawl and debugging felt endless. After all that the thing just crashed and became a Slither io useless.