Applying Deep Learning, LLMs and AI Agents in Hong Kong Commercial Buildings

Conventional FDD rules are hand-designed by engineers, project by project, and hard-coded into building systems — costly to author, difficult to enumerate and change, and unable to adapt when sensors degrade, operating modes shift, or instruments contradict one another. This session presents a pilot implementation of LLM-based adaptive FDD for air handling units in a large Hong Kong commercial facility — tracing the journey from a naive LLM approach to a grounded, continuously improving architecture.