Domain-Specific LLM Aids Residential Energy Retrofit Decisions

A study published on arXiv reports a domain-specific large language model (LLM) designed to help homeowners make informed residential energy retrofit decisions. The model uses parameter-efficient low-rank adaptation (LoRA) fine-tuning on a corpus from 536,416 U.S. residential building prototypes, based on physics-based energy simulations and techno-economic calculations. It evaluates nine major retrofit categories including envelope upgrades, HVAC systems, and renewable energy installations. The LLM relies solely on homeowner-accessible natural-language descriptions such as building age, size, and location. Validations against physics-grounded benchmarks show consistent performance. The research addresses the expertise gap that stalls retrofit initiation due to homeowners' lack of technical literacy.