Inference time LLM alignment in single and multidomain preference spectrum