Luca Pasquarelli
Junior Researcher in the Energy Efficient Buildings group – Institute for Renewable Energy, Eurac Research in Bolzano, Italy. Degree in Computer Science and Engineering. Former member of the Tech Transfer Digital at NOI. Particularly interested in Generative AI, Natural Language Processing and all the overlapping domains between tech, society and humanities.
Session
Energy Management and Information Systems (EMIS) in buildings often face significant challenges related to scalability and interoperability. These limitations stem primarily from the absence of standardized metadata models, which creates barriers when implementing solutions across multiple buildings with different naming conventions and system configurations.
The Brick ontology offers a promising solution as an open-source framework designed to standardize the description of physical and logical building entities. However, creating ontology-based metadata models requires specialized knowledge in both building systems and semantic modeling—a rare combination of expertise that limits widespread adoption.
This talk introduces BrickLLM, an innovative Python library (EURAC, Politecnico di Torino) that leverages Large Language Models (LLMs) to automatically generate Resource Description Framework (RDF) graphs using the Brick ontology. BrickLLM transforms natural language descriptions of buildings and their energy systems into machine-readable, standardized metadata models, dramatically reducing the time, cost, and expertise required for this critical task.
Key features of BrickLLM include:
1. Support for both cloud-based LLM APIs (OpenAI, Anthropic, Fireworks AI) and local open-source models (LLaMA 2, LLaMA 3)
2. Two workflow options: a streamlined "instruct" workflow for fine-tuned models and a more structured workflow for general-purpose LLMs
3. Built-in validation against Brick's SHACL shapes to ensure semantic correctness
4. Modular architecture based on LangChain and LangGraph for extensibility
The presentation will demonstrate how BrickLLM enables users without specialized knowledge to generate comprehensive, standards-compliant RDF representations of buildings and their systems through simple natural language descriptions. We'll explore the practical applications of this technology, including:
- Facilitating portable EMIS applications across different buildings
- Enabling more efficient energy optimization strategies
- Supporting fault detection and diagnosis systems
- Enhancing interoperability between different building management platforms
By democratizing access to ontology-based metadata modeling, BrickLLM addresses core challenges in the open data ecosystem for smart buildings and energy management. The library's open-source nature encourages collaboration and further research, potentially leading to optimized workflows for semantic metadata model generation.
We'll also discuss how this approach aligns with broader initiatives to standardize and democratize data across different domains, making it particularly relevant to the Open Data Hub Day's mission of promoting best practices in data management and utilization.