Use these documents to help inform the creation of a Google Cloud hybrid search architecture and deployment:
Architecture and design guidance
- Secure agent interactions with MCP Toolbox for Databases: Information on configuring security, connection management, and agent interaction boundaries using MCP Toolbox for Databases.
- RAG infrastructure for generative AI using Agent Platform and AlloyDB for PostgreSQL: An architecture that stores vector embeddings alongside your operational data in a fully managed database like AlloyDB for PostgreSQL.
- Best practices for tuning ScaNN index in AlloyDB: Best practices for tuning ScaNN index parameters like num_leaves, num_leaves_to_search, and pre_reordering_num_neighbors for high recall and fast query performance.
- Choose a connectivity option for AlloyDB: Compares AlloyDB connectivity methods, including AlloyDB Auth Proxy, Private Service Connect (PSC), and direct VPC peering.
- AlloyDB security best practices: Recommendations to secure AlloyDB instances using IAM database authentication, Private Service Connect, and VPC perimeters.
Deployment
- Hybrid search on Cloud Run codelab: A step-by-step hands-on tutorial for building and deploying a hybrid search application using AlloyDB, Gemini Enterprise Agent Platform, and Cloud Run.
- Run hybrid vector similarity search in AlloyDB: Guidance to construct and execute hybrid search queries combining vector similarity ordering with relational SQL filtering in AlloyDB
- Evaluate semantic queries using AI operators in AlloyDB: Guidance to use in-database AI functions such as ml_predict_row and ai.rank for semantic reranking and LLM validation.
- Cloud Run Direct VPC Egress: Instructions to set up Direct VPC Egress for Cloud Run services to access private resources like AlloyDB within a VPC network.