Vector Annotation Databases (VAD)

An Architecture for Auditable Semantic Retrieval: Proposes a solution to opaque vector retrieval in AI memory systems: the Vector Annotation Database (VAD), in which deterministic metadata forms the retrieval surface and embeddings function as swappable annotation layers—yielding auditability, cross-model portability, and regulatory compliance for AI retrieval systems.