Academy of Interdisciplinary Studies Division of Integrative Systems and Design 257 AIoT Empowered Smart Traffic Systems Supervisor: SONG Shenghui / ISD Student: FENG Shiqi / ISD WONG Chun Ka / ISD ZHANG Yunxin / ISD Course: UROP 1100, Summer This report presents an AIoT-enabled smart traffic system capable of real-time traffic monitoring and adaptive control. The system comprises three key components: a smart car, modular smart tiles, and a backend server for traffic control and real-time data processing. By integrating artificial intelligence (AI), the prototype offers a tangible platform for exploring IoT networking, sensor fusion, and real-time decisionmaking. Leveraging cloud connectivity (Firebase), it demonstrates how IoT devices communicate, dynamically respond to environmental changes, and coordinate actions—simulating real-world scenarios such as vehicle-to-infrastructure interaction such as cars responding to traffic signals. Future applications could extend to research or education, serving as interactive teaching material to model real-time road interactions (e.g., car-to-traffic-light communication) in a controlled environment. LLM for Networking Supervisor: SONG Shenghui / ISD Student: XU Borong / CPEG Course: UROP 2100, Fall In the previous stages, the LLM was tested with the Chinese card game “Landlord”. Despite the earlier findings with ChatGLM, several improvements have occurred since the last report. The game system has been upgraded with better prompt engineering feasibility and also implemented with OpenAI API to have a wider selection of LLM models. This report introduced a new method to evaluate the LLM in understanding gaming to compare the performance between different models. And will use the method to evaluate different prompt engineering approaches. Among several popular prompts that potentially increase model performance, this report discusses several prompts that would influence the performance of the model in card games. In the end, this report will discuss further improvements for gaming LLM in the next stage. LLM for Networking Supervisor: SONG Shenghui / ISD Student: SHEN Yuming / ISD Course: UROP 1100, Fall UROP 2100, Summer This report presents Event Pulse, an AI-driven network optimization system that analyzes social media data to predict crowd distribution and dynamically allocate network resources. Building upon previous UROP research on LLM applications in networking, we implemented enhanced algorithms for region-based impact analysis and priority aware resource mapping. The system demonstrates 92% prediction accuracy and 35% network utilization improvement in simulations. Current limitations include synthetic data dependency and API rate constraints. Future work will integrate real-time social media feeds and edge computing deployment. This project establishes a foundation for semantic-aware network resource management using generative AI technologies.
RkJQdWJsaXNoZXIy NDk5Njg=