The theory
The semantic cube: meaning as geometry
At Semantic Geometry we do not store text or probabilities: we project every concept to a point in a continuous [-1, 1]³ cube. The distance between two points is how close their meanings are, and a claim is verified by walking the graph — not by guessing the next token.
A continuum, not a table
Each axis of the cube is an IEEE 754 double: about 9.2·10¹⁸ values per dimension and on the order of 10⁵⁶ unique points overall. That is more room than there are atoms on Earth — far beyond any human vocabulary; today’s concepts fill an infinitesimal fraction of the cube.
The AI proposes, silicon verifies
A language model can draft; the cube checks. Every critical claim is anchored to coordinates and to an auditable graph, and the verdict resolves in sub-milliseconds via grounding: zero tokens to verify, zero GPU, zero network.
Deterministic and auditable
The same claim against the same ontology yields the same result on any machine. The knowledge base is sealed with a Merkle ledger (BLAKE3) that detects any corruption or tampering, and the engine runs offline at the edge (~15 W).
Deterministic AI from Rancagua, for Chile and Latin America
Semantic Geometry Software & AI SpA is a deterministic artificial intelligence company based in Rancagua, O’Higgins Region, Chile. We work on high-performance software architecture, data pipelines and deterministic neuro-symbolic AI. The thesis is simple: a deterministic AI from Rancagua can give Chile —and all of Latin America— the ability to validate language without depending on a foreign black box. From the mining edge of O’Higgins to any 15 W device, the deterministic AI from Chile we build verifies in silicon, locally, without giving the data away.
Want to apply this to your domain?
Tax, labor, automotive and credit verticals already run on the cube. Tell us your case.
Talk to us