Schema-Aware Query Routing with Temperature Classification and AI Workload Detection

We present a schema-aware query routing system for database proxies that automatically classifies tables by data temperature (HOT/WARM/COLD based on inter-query arrival times) and workload type (OLTP/OLAP/Vector based on query structure), detects AI-specific workloads (RAG retrieval, embedding batch inserts, agent conversation patterns), and routes queries to optimal backend nodes using a learning classifier that adapts table classifications from observed access patterns without manual configuration. The system includes automatic schema discovery via information_schema queries, shard key detection from index analysis, and relationship-aware routing that co-locates related table queries. While database proxies with routing exist (Vitess vschema, ProxySQL regex rules, Citus distribution columns), none combine automatic temperature classification, AI workload detection, and an adaptive learning classifier in a single system.