Enterprises are deploying agentic AI throughout the software development lifecycle and need the behavioral logic of the systems they act on, not stakeholder documentation. To obtain this logic, a seven‑agent reverse‑engineering pipeline was built, processing 241 repositories across 18 services and producing thousands of artifacts that describe scope, journeys, rules, and acceptance criteria.
Recovery alone does not guarantee usability. Entities such as offer, eligibility, and activation appear repeatedly across services with a shared vocabulary but distinct operational semantics. A single end‑to‑end journey can involve over a thousand near‑identical rules that must not be conflated. Flat vector retrieval returns an identical rule from the wrong service with high similarity and provides no error signal. A traditional content‑level ontology does not exist, and relating every occurrence of recurring entities would cost more than the engagement and would not survive a release cycle.
The solution places the ontology at the artifact level. Each machine‑generated artifact retains provenance information—producer, component, service, domain, and type—allowing disambiguation before the content is read. The knowledge layer consists of a self‑describing catalog, a sparse metadata registry, and a navigation hypergraph defined over artifacts rather than extracted entities. Queries resolve to artifact identifiers, which constrain vector search so that competing rules from other services are absent rather than merely outranked, making disambiguation structural instead of statistical.
Built in under a week, new services can be admitted by convention. The layer serves SDLC agents, developers, quality engineers, and analysts. It has supported root‑cause analysis and resolution of over 100 production defects with developer verification and has attributed 35 release regression failures to specific rules and services.
Review: By introducing a structured ontology at the artifact level, the approach achieves precise differentiation of massive, similar rules, offering a scalable and efficient knowledge organization for enterprise‑grade RAG systems.