Enterprise GenAI Agent Architectures on Google Cloud
Verified outcome — Three architectures I specified and verified on Google Cloud Run — grounded RAG with Vector Search, BigQuery SQL reasoning over an MCP server, and sandboxed Python execution. Selected as a featured Hack2Skill APAC submission.
- My verified role
- Cloud AI Architect & Agent Director
- Classification
- Feature
- Primary tools
- Google Cloud Run, Vertex AI (Gemini 2.5/3.6 Flash), Google ADK
- Public evidence
- Demo videoSource code
Source codeThe Problem
Enterprises require proven, secure agentic patterns to connect LLMs to unstructured knowledge bases, big data warehouses, and automated operational workflows without security risks.
My Role
Cloud AI Architect & Agent Director
What I Personally Directed
Multi-pattern architecture design, Model Context Protocol (MCP) integration, Cloud Run Sandboxes configuration, least-privilege IAM security, and human-in-the-loop governance.
The Solution
A production-grade trilogy of AI Agent architectures featuring: (1) Grounded RAG with Vector Search, (2) Autonomous BigQuery SQL reasoning via MCP Server, and (3) Dynamic Python execution inside Cloud Run Micro-Sandboxes with Google Sheets API and WebSockets.
Verified Outcome
Three architectures I specified and verified on Google Cloud Run — grounded RAG with Vector Search, BigQuery SQL reasoning over an MCP server, and sandboxed Python execution. Selected as a featured Hack2Skill APAC submission.