Rare diseases affect approximately 300 million people worldwide, and over 95% of rare diseases lack approved treatments. The diagnostic process for rare diseases is often prolonged and error-prone, with patients receiving several misdiagnoses and consulting multiple specialists before receiving an accurate diagnosis. The Radiant project aims to mitigate these challenges by providing a reliable, efficient, and context-aware information retrieval system with a user-friendly chat interface. Leveraging large-language models (LLMs), retrieval-augmented generation (RAG), and a domain-specific vector database for efficient query-time retrieval of relevant source information, Radiant generates responses tailored to rare disease queries. Transparency is enhanced by annotating responses with clickable primary source references. Radiant is intended to support healthcare providers, patients, and researchers in exploring disease mechanisms, genetic underpinnings, and potential therapeutic strategies. Radiant’s design minimizes the risk of inaccuracies by reranking retrieved documents for relevance before integrating them into the response generation process. This approach reduces the reliance on the inherent limitations of standalone LLMs, such as hallucinations and overgeneralization. Future improvements will include expanding the coverage of diseases, refining the underlying algorithms, and integrating patient-specific data to enhance clinical relevance and usability.