I’ve been exploring how we might treat knowledge as a geometric structure instead of a collection of embeddings. KAG (Knowledge as Geometry) positions facts, classes, and instances in a shared coordinate space — making retrieval spatial, about distance, density, and context rather than simple vector similarity. It’s an early-stage research prototype — not production-ready, but meant to spark discussion. I’d love feedback from people working on retrieval, embeddings, or conceptual modeling.