I have reworked a lot of code i built myself with AVX2 and Neon, to become much better, more maintainable and faster too, with LLMs.