Home

Human–AI Co-Design for Urban and Educational Systems:
Multi-Agent Participatory Platforms for Planning and School Choice

University of Florida, Urban Tech & Design Lab, (Supervisor: Zhaoxi Zhang)

American Association of Geographers (AAG) Annual Meeting (2026)


       Building on my concern with cognitive accessibility, this project asks a deeper question: how do people develop (or withdraw) epistemic trust in intelligent systems during collective decision-making? Instead of using AI only as a generative tool, we treat agents as knowledge mediators that surface constraints, trade-offs, and precedents inside the design dialogue itself.
       
       I built the technical architecture for a multi-user platform where humans and AI agents co-create urban design proposals and explore school-choice options. Every design revision and conversational turn is logged, allowing us to study how people negotiate recommendations, challenge system suggestions, and gradually form mental models of “what the AI is doing.”

  • AI-Mediated Urban Co-Design PlatformAI: Developed a browser-based, multi-user environment where participants and AI agents iteratively revise streetscape and neighborhood layouts, with agents explaining zoning, environmental, and equity constraints as part of the conversation.

  • Participatory AI as Knowledge Mediator: Designed multi-agent behaviors that prioritize constraint explanation, comparison of alternatives, and clarification of assumptions—shifting AI from a one-shot generator into a partner that supports reasoned trade-off exploration.

  • Educational Decision Support Interface: Built spatial data pipelines and an interactive advisory front end for K–12 school selection, integrating a RAG-based agent that provides transparent, constraint-aware recommendations grounded in distance, program offerings, capacity, and policy rules.

  • Dialogue and Revision Logging (Ongoing): Implemented a logging infrastructure that records dialogue traces, design versions, and interaction timelines, enabling analysis of sensemaking strategies, trust calibration, and reliance patterns in both civic co-design and school advisory settings.

I want to study power and trust in sociotechnical systems, by leverageing these traces to examine how interface framing and agent behavior redistribute interpretive power between institutions, communities, and AI systems—informing design principles for more legible and negotiable human–AI collaboration.