Frame 1 of 4
ai·rete·rag separates decision logic from explanation with Rete engine and RAG LLM
ai·rete·rag runs a deterministic Rete rule engine first to evaluate YAML rules and produce consistent verdicts with salience-based conflict resolution. Then, a retrieval-augmented generation LLM retrieves passages from policy documents and writes plain-English explanations citing those passages without changing the verdict. The system supports nested rules, forward chaining, and audit mode recording every evaluated rule with snapshots for replay. Non-technical authors can create rules via a visual editor or get LLM-drafted rules from policy documents, with drafts requiring review.