Researchers introduce Ready2Blend, a method that combines natural-language instructions with learned alignment for LLMs. It uses AlignFormer to map requirements to alignment prompts and enables inference-time blending and reweighting. Ready2Blend achieves competitive results with post-training-based alignment methods while requiring less training time.