Anthos Insight

Gemini-powered agents on GKE extend Bank of Anthos via MCP: detect anomalies, coach spending, and execute safe transfers—external AI services, no core code changes.

Comment

Inspiration

When GKE turned 10, we wanted to celebrate by showing how agentic AI could upgrade a flagship demo like Bank of Anthos without rewriting its code. Inspired by real banking trends—AI assistants, fraud prevention, and personalized finance—we set out to prove that cutting-edge AI can be layered onto production microservices safely and intelligently.

What it does

Anthos Insight adds a containerized AI layer beside Bank of Anthos. New agents provide:

Intelligent Banking Assistant – natural language banking powered by Vertex AI Gemini

Advanced Fraud Detection – anomaly detection with explainable insights

Personalized Financial Coach – spending analysis and savings recommendations

Secure Transfers – guided fund movements with confirmation checks

Real-time Insights – categorized spending breakdowns and analysis

How we built it

Architecture principle: zero-touch extension, no edits to the Bank of Anthos core.

GKE Autopilot for container orchestration and scaling

Vertex AI Gemini 2.5 Flash for reasoning and conversation memory

MCP server to expose Bank of Anthos APIs as agent tools

Agent Gateway (FastAPI + ADK) to orchestrate multi-agent workflows

Specialized agents for fraud detection, coaching, and support tasks

Enhanced UI with chat, insights, and confirmation steps

Security via Workload Identity, JWT handling, and NetworkPolicies

Challenges we faced

Zero-touch integration – solved with an MCP façade layer

Complex banking data and JWT authentication – addressed with secure adapters

Context injection for Gemini – managed with prompt engineering and streaming data

Multi-agent orchestration – implemented intent-based routing with ADK

Resource limits – optimized through GKE Autopilot autoscaling

Accomplishments

Production-grade, containerized overlay on Bank of Anthos

Real banking conversations powered by Gemini

Enterprise-level security with Workload Identity and policies

Professional UI enhancements aligned to Bank of Anthos style

Fast responses and efficient scaling on GKE

What we learned

Advanced usage of GKE Autopilot and workload identity patterns

Practical agent engineering with MCP, A2A, and Gemini

How to extend existing microservices through API-first design

Insights into banking data structures, fraud detection, and UX design

What’s next

Deeper integration with the Agent Development Kit (ADK)

Expansion to mobile and multi-language support

Advanced security features including biometric authentication

Open-source release as an extension framework for Bank of Anthos

Built With

  • asyncio)
  • configmaps
  • container-registry
  • html/css-google-cloud:-gke-autopilot
  • javascript-(react)
  • jwt-authentication
  • kubernetes-manifests
  • languages-&-frameworks:-python-(fastapi
  • loadbalancer
  • mcp
  • multi-agent-workflows-infrastructure-&-devops:-docker
  • real-time
  • secrets-integration:-bank-of-anthos-apis
  • vertex-ai-gemini-2.5-flash
  • workload-identity-agentic-ai:-adk
Share this project:

Updates