Scaling the new generation of Energy Management and Information Systems (EMIS) across different buildings and energy systems presents significant challenges, mainly due to standardization and interoperability issues. Semantic metadata schemas can effectively address these challenges through data standardization across building sensor networks, and the abstraction of buildings and energy systems with graph structures. However, developing metadata models for each building is time-consuming and requires specific domain expertise. Generative Artificial Intelligence (GenAI) offers a potential solution specifically utilizing Large Language Models (LLMs) to generate metadata models from prompts describing buildings, sensors, and energy systems. However, traditional LLMs often struggle with accuracy in specialized tasks, leading to the exploration of an Objective-Driven approach. This approach emphasizes the use of smaller, fine-tuned LLMs that are optimized for specific tasks, making them more accurate and efficient while also being lightweight enough to run on local machines with limited computational resources. The study specifically evaluates a specialized LLM, “llama3.1-brick”, which has been fine-tuned on hundreds of metadata models based on the Brick ontology. The model performance is assessed through a validation pipeline designed to assess its ability to generate semantically accurate metadata models from textual descriptions. The results of a real-world case study prove the capability of the model in successfully generating accurate metadata models for several store buildings. The work demonstrates the potential of the proposed approach in effectively addressing the challenges of scaling EMIS offering a more efficient and accessible solution to metadata model generation.