Short Title: AzurePostgreSQL Profiler: AI-Driven SKU Sizing Using Customer Metadata Without Performance Traces Abstract: We present AzurePostgreSQL Profiler, a novel AI-powered system for automating initial SKU selection in Azure PostgreSQL Database deployments—without requiring any prior workload telemetry. Traditional approaches depend on historical performance traces, rendering them impractical for newly provisioned databases. Our system introduces a three-phase recommendation engine: first, it right-sizes existing PostgreSQL instances by modeling slack and throttling trade-offs; second, it uses hierarchical and target-encoding models to recommend capacities based solely on customer profile metadata (e.g., industry, subscription tier, resource group); and third, it continuously personalizes recommendations via a feedback loop that ingests customer-reported incidents and scaling actions. Evaluated on production Azure PostgreSQL flexible server workloads, our system reduces wasted capacity by >60% compared to user-selected defaults while maintaining performance thresholds. The framework is fully implemented in Azure’s provisioning pipeline, providing explainable, low-latency SKU recommendations that adapt to evolving customer preferences.