During the seminar, the results of a research project focused on generating realistic Zigbee network traffic using a Large Language Model (LLM), specifically OpenAI's GPT-4.1, will be presented. Unlike traditional rule-based or statistical approaches, the proposed method extracts representative packet samples from real Zigbee traffic.

It incorporates sample-based learning, human-in-the-loop feedback, and prompt engineering. The presentation will include two key experiments:

One focused on generating unidirectional device-to-hub communication, and another on bidirectional traffic to capture realistic interaction patterns. The effectiveness of the approach is evaluated using the Jensen-Shannon Divergence metric, demonstrating that GPT-4.1 can produce semantically meaningful and protocol-compliant synthetic traffic. This work highlights the potential of LLM-based traffic generation for enhancing IoT simulation, testing, and security evaluation.

During the seminar, the results of a research project focused on generating realistic Zigbee network traffic using a Large Language Model (LLM), specifically OpenAI's GPT-4.1, will be presented. Unlike traditional rule-based or statistical approaches, the proposed method extracts representative packet samples from real Zigbee traffic.                   It incorporates sample-based learning, human-in-the-loop feedback, and prompt engineering. The presentation will include two key experiments:

One focused on generating unidirectional device-to-hub communication, and another on bidirectional traffic to capture realistic interaction patterns. The effectiveness of the approach is evaluated using the Jensen-Shannon Divergence metric, demonstrating that GPT-4.1 can produce semantically meaningful and protocol-compliant synthetic traffic. This work highlights the potential of LLM-based traffic generation for enhancing IoT simulation, testing, and security evaluation.