In today’s artificial intelligence (AI) driven world, modern systems communicate with people with diverse backgrounds and skillsets. Hence, for this human-machine interaction to be more meaningful, it is important for machines to be aware of the context and user expertise. Therefore, this study proposes an agentic AI profiler that uses four levels: Novice, Basic, Advanced, and Expert to classify natural language responses. The system was built with a modular layered architecture and LLaMA v3.1 (8B) with components for text preprocessing, scoring, aggregation, and classification. The system evaluation was done in 2 phases: static and dynamic. The static phase analyzed pre-recorded transcripts from 82 participants. Similarly, the dynamic phase evaluated 402 live interviews conducted with an agentic AI interviewer. In both phases, participants’ self-ratings were compared with the profiler’s rating. However, during the dynamic phase, participants’ expertise was evaluated after every response, not at the end of the interview as in the static phase. Across domains, 83%–97% of profiler evaluations matched participants’ self-evaluations. The remaining differences came from participants overrating or underrating themselves, unclear responses, and occasional LLM misinterpretations of nuanced expertise.