Milan, June 2025 — At CLIMA World Congress 2025, the MODERATE project took center stage with the presentation of a pioneering research poster by Rocco Giudice from the BAEDA Lab at Politecnico di Torino, developed in close collaboration with Eurac Research.
The work introduces llama3.1-brick, a lightweight and specialized Large Language Model (LLM) fine-tuned to generate Brick-compliant semantic metadata models from natural language building descriptions.
Why it matters
Creating consistent and structured semantic metadata is essential for enabling Energy Management and Information Systems (EMIS) to scale across real-world applications. However, generating these models is typically time-intensive and demands expert knowledge. llama3.1-brick addresses this bottleneck by enabling:
-
Automatic RDF graph generation from textual inputs
-
Local deployment, ensuring privacy-preserving and cost-efficient use
-
Robust performance across diverse building typologies
Technical Highlights
-
Uses parameter-efficient fine-tuning (QLoRA) and instruction tuning
-
Fine-tuned on Brick-based metadata to ensure structural and semantic validity
-
Tested on 16 real retail buildings, demonstrating high success rates in just a few iterations
Real-World Impact
This model offers a scalable, flexible, and privacy-conscious solution for building digitalization — especially valuable when direct sharing of real building data is restricted. The outcomes signal a major step forward for AI-assisted interoperability in the building and energy sector.
A Collaborative Effort
The research is part of the MODERATE Horizon Europe initiative and was made possible thanks to a cross-institutional collaboration, including:
-
Rocco Giudice, Prof. Alfonso Capozzoli, Dr. Marco Savino Piscitelli (Politecnico di Torino)
-
Marco Perini, Daniele Antonucci, Dr. Cristian Pozza (EURAC Research)
Stay tuned as MODERATE continues to push the boundaries of data-driven interoperability in energy systems.
