Brick-DICL: Dynamic In-Context Learning for Automated Brick Schema Classification
Researchers have developed Brick-DICL, a novel two-stage framework designed to automate the classification of building management system points into the standardized Brick schema. This approach addresses challenges such as the vast number of Brick classes, limited domain knowledge in LLMs, and the need for manual verification. Brick-DICL utilizes metadata-RAG and class-RAG components to enhance LLM knowledge and narrow down classification options, while a multi-LLM filtering mechanism flags uncertain predictions for human review. The framework demonstrates significant accuracy improvements and reduces manual effort, accelerating the integration of digital building management systems. AI
IMPACT This research could streamline the integration of diverse building management systems, paving the way for more efficient and interoperable smart buildings.