Embed a LLM at the inference stage to turn your existing ML inference pipeline into an ensemble and gain some low-hanging precision recall improvements.