HyEm: Query-Adaptive Hyperbolic Retrieval for Biomedical Ontologies via Euclidean Vector Indexing

HyEm targets a very specific technical bottleneck in RAG: the mismatch between hierarchical data (biomedical taxonomies) and the Euclidean geometry used by modern vector databases. While the approach is scientifically sound and addresses a real-world problem for specialized Bio-AI applications, its defensibility is low (score 3) due to its status as an academic reference implementation with zero stars and minimal community traction. The primary moat is the specific mathematical mapping between hyperbolic and Euclidean spaces for query-adaptive retrieval, but this is a 'math moat' rather than a 'data' or 'network' moat. Frontier labs like Google (Med-PaLM) or specialized players like Recursion/BenevolentAI are the primary competitors; however, the biggest risk is platform domination from vector database providers (Pinecone, Milvus, Weaviate). If hierarchical retrieval proves essential, these platforms will likely implement native hyperbolic distance metrics (e.g., Poincaré distance), rendering 'wrapper' algorithms like HyEm obsolete within 1-2 years. Its current value lies in being a modular algorithm for specialized RAG pipelines where hierarchy matters more than raw semantic similarity.