Rare diseases affect fewer than 1 in 10,000 individuals but collectively lead to significant, lifelong disabilities. In Sweden, approximately 500,000 people—nearly 5\% of the population—live with one of the 6,000 to 8,000 identified rare conditions. These patients often endure a "diagnostic odyssey," waiting years for an accurate diagnosis. Closing this diagnostic gap is a cornerstone of Sweden’s national life science strategy, necessitating scalable, digital solutions to integrate fragmented clinical data across the 21 healthcare regions. The \textbf{aim} of this project is to develop specialized Large Language Models (LLMs) to accelerate the diagnosis of rare diseases. We propose a modular AI framework that integrates heterogeneous Swedish clinical data—ranging from electronic health records (EHRs) to specialized genomic reports—without the computational overhead of monolithic models. Specifically, the project seeks to: \begin{itemize} \item Engineer modular ``expert modules'' that capture rare-disease-specific knowledge. \item Integrate these modules into a global diagnostic system to democratize specialist expertise across all Swedish healthcare regions. \item Enable sustainable Precision Medicine by reducing the socio-economic burden on the national healthcare system. \end{itemize} By providing a computationally efficient and scalable path toward jämlik vård (equal care), this research ensures that high-quality diagnostic insights are accessible regardless of a patient's geographic location within Sweden.