crit

What if building a tailored, high-performance inference engine took hours instead of weeks?