Recent breakthroughs in Large Language Models (LLMs) have redefined the landscape of human-computer interaction (HCI), particularly in intelligent dialogue systems and automated decision-making. In recruitment, interviews remain resource-intensive due to their demand for context-aware interaction and reliable candidate evaluation. Very few existing work provide fully integrated, multi-turn simulations that emulate real recruiter–candidate dialogues. We introduce MirrorMind, an LLM-powered simulation framework, to record multi-turn interviews through dual-agent modeling. Unlike Q&A systems tailored to narrow technical domains, MirrorMind generalizes across industries and roles, enabling broader applicability. Using a dual-agent architecture enhanced by retrieval-augmented generation (RAG), MirrorMind enables realistic role-play between the interviewer and the interviewee. Instead of submitting the interviewers’ CV to the real interviewee, MirrorMind only upload CVs to the LLM data base, which can protect the interviewers’ privacy. We validate it in both local and API settings, demonstrating its practicality for automated recruitment, candidate training, and performance profiling.