LLM-based tools for the automated implementation and energy optimization of building automation systems (SAGE)
Third party funded individual grant
Acronym: SAGE
Start date : 01.09.2026
End date : 31.08.2029
Project details
Scientific Abstract
The SAGE research project aims to significantly improve the efficiency and sustainability of building
automation through the systematic use of Large Language Models. The existing German building stock
harbors considerable savings potential that has so far remained untapped due to outdated automation
systems and inadequate data structures. A central obstacle in this regard is the so-called "onboarding
problem": building documentation in existing structures typically exists in heterogeneous formats and is
distributed across various sources such as Excel spreadsheets, CAD drawings, and technical manuals,
without any central, structured database — making the implementation of advanced control strategies
considerably more difficult.
SAGE addresses this challenge with a novel, LLM-based approach to the automated processing of
heterogeneous building documentation. Using Retrieval-Augmented Generation and GraphRAG
techniques, unstructured sources such as sensor lists, as-built drawings, and technical documentation are
transformed into machine-readable semantic knowledge graphs. These form the basis for subsequent LLMassisted
building topology identification and the automated design of suitable control concepts. The high
degree of automation in the developed methodology demonstrates a scalable path to unlocking significant
energy savings in existing buildings through optimized system operation with minimal manual effort.
The developed methods are validated at the Bosch site in Renningen under real operating conditions. All
software components will be made available to the professional community as open-source contributions
and disseminated through an industry workshop.
