Automated interviewers and conversational agents are increasingly common in research, recruitment, customer service, and education. However, many current systems depend on fixed question sequences and provide only limited context-based adaptation, which can make interactions feel rigid and less responsive to individual respondents.
The paper introduces an evidence-traceable dynamic interviewer architecture built around local large language models. Rather than following a static script, the system adapts its questions to the interviewee's expertise, and each question can be traced back to the evidence that prompted it. This combination of adaptability and traceability is designed to make qualitative interviews more flexible while keeping the reasoning behind each question transparent.
Because the architecture runs on local LLMs, it also offers potential privacy benefits by avoiding cloud-based processing. The abstract does not include evaluation results, so the practical effectiveness of the approach remains to be seen, but the design directly addresses known limitations in current conversational interviewing systems.