Comment by kiyeonjeon

Love the "no LLM calls" approach. Scoring prompts in <1ms locally is exactly the right tradeoff. Most tools overcomplicate this.