How we reduced AI inference costs by 60% without sacrificing accuracy

· Dev.to