GenAI-Powered Multi-Model Query Optimization: Bridging SQL, JSON, and Vector Search in Oracle-MongoDB Ecosystems

  • Unique Paper ID: 180155
  • Volume: 12
  • Issue: 1
  • PageNo: 347-353
  • Abstract:
  • This paper introduces a neural architecture for cross-database query optimization, leveraging generative AI to bridge SQL (Oracle), JSON (MongoDB), and vector search operations. Our transformer-based model, trained on execution plans from Oracle Autonomous Database and MongoDB Atlas, demonstrates 40–65% latency reduction for hybrid queries combining SQL joins, JSON aggregations, and semantic vector searches compared to manual optimization. The system dynamically selects execution pathways using real-time workload analysis inspired by Gemini for Databases recommendations [1], while resolving schema mismatches through AI-driven JSON-to-relational mapping. By integrating Oracle 23ai’s vector indexing with MongoDB’s native vector search, we achieve sub-100ms response times for complex analytical workloads. Security constraints from Oracle’s ethical AI governance are preserved through differential privacy in query translation. Evaluation shows 2.7× improved resource utilization in hybrid cloud deployments, with vector search recall rates exceeding 92% on TPC- H benchmarks. This approach enables intent-driven data interaction patterns while abstracting database heterogeneity, advancing GenAI-enhanced data mesh architectures.

Related Articles