Neuromuscular diseases (NMDs) are a heterogeneous group of rare disorders that significantly impair motor function and quality of life. Their clinical and genetic variability makes systematic study and knowledge integration particularly challenging. We present a scalable, automated framework that combines natural language processing and network analysis to uncover hidden relationships among NMDs. Using Sentence Transformer models, we identified NMDs and their phenotypes in open-access documents. Then, we use this data to build NMD networks to infer new relations using embedding techniques and clustering approaches structuring a document corpus for 328 NMDs basis on disease type and associated to phenotypes and genes. These findings show the value of combining language models and network embeddings for large-scale rare disease analysis.