Llamas on the Edge: building an AI personalization engine with edge workers