A three-tier hierarchical RCA engine that progressively refines unstructured support tickets through technical anomaly extraction, business keyword clustering, and contextual AI vector discovery. Tier 1 scans SAP error signatures and applies agglomerative sub-clustering; Tier 2 locks tickets to predefined business objects (PO, GRN, SES, PR, STO, ASN) using cosine similarity; Tier 3 processes long-tail conversational tickets via BERTopic, UMAP, and HDBSCAN in compressed vector space. A unified quality gate applies centroid calculation, hard cosine pruning (0.85 threshold), few-shot ITIL prompting via Phi-3, and programmatic deduplication across all tiers. This architecture solves the challenge of automating RCA categorization at scale while maintaining business context and technical precision across heterogeneous ticket sources.