Closing the Expertise Gap in Residential Building Energy Retrofits: A Domain-Specific LLM for Informed Decision-Making

A research team from Michigan State University and Penn State has developed a specialized large language model (LLM) designed to bridge the expertise gap in residential energy retrofitting. The model, detailed in a new arXiv preprint, addresses a critical barrier to home decarbonization: homeowners lack the technical knowledge to assess which energy upgrades—like new windows, insulation, or heat pumps—offer the best return on investment and environmental impact. By translating complex engineering and economic calculations into accessible advice, this AI tool aims to democratize access to expert-level retrofit planning.

The model's core strength lies in its training on a massive dataset of 536,416 physics-based energy simulations covering nine major retrofit categories across U.S. residential prototypes. Using an efficient fine-tuning method called Low-Rank Adaptation (LoRA), the LLM learns to map simple homeowner descriptions of a dwelling (e.g., size, age, location) to optimal retrofit selections and their projected outcomes. The results are striking: evaluation shows it recommends the retrofit with the highest CO2 reduction within its top three choices 98.9% of the time and identifies the option with the shortest discounted payback period 93.3% of the time. Furthermore, fine-tuning reduced CO2 prediction error by an order of magnitude and significantly cut errors for energy use and cost estimates. The model also maintains robust performance even with incomplete homeowner input, making it a practical tool for accelerating residential decarbonization at scale.