Built a hybrid log classification system using three sequential layers: regex rules for fast pattern matching, a machine learning classifier for structured patterns, and an LLM fallback for ambiguous cases. Designed for high-volume log environments where single-method classification fails