Back to blog
Architecture4 min readDraft date
Separating Ontology Metadata from Model Artifacts
Queryable metadata and large ML artifacts have different jobs. Treating them separately keeps the engine inspectable without exposing storage internals.
Ontology metadata and model artifacts have different access patterns. Metadata needs to be queryable for decisions, inspection, and future workload context. Artifacts need storage boundaries suited to datasets, checkpoints, generated outputs, and packaged models.
Aero describes this separation conceptually. The public architecture avoids deployment topology, storage identifiers, schema details, and production configuration.