Meet Your AI Bankers: Next-Gen Bank with AI Employees!

Inspiration

Banking reimagined for the digital age. We wanted to eliminate the frustration of waiting in lines and limited banking hours by creating a bank that's open 24/7 with AI employees ready to help instantly. The inspiration came from transforming the classic Bank of Anthos into a next-generation, AI-powered banking experience.

What it does

AI-powered banking platform provides two main AI agents that work together:

Customer Service AI Agent: Users can chat with an AI assistant to check balances, send money, manage contacts, and get information about promotions they're eligible for.

Promotion Management AI Agent: Automatically creates custom promotions (like "get a bonus when you deposit $1000"), monitors user transactions to check qualification, and automatically credits rewards to accounts when users meet promotion criteria.

How we built it

• Google Agent Development Kit (ADK) with Gemini 2.5 Flash model for conversational AI
• Agent-to-Agent (A2A) Protocol allowing our CS and Promotion agents to communicate
• Model Context Protocol (MCP) to standardize banking actions as tools that AI agents can use
• Google Kubernetes Engine (GKE) for cloud deployment and scaling
• Vertex AI the brain behind the agents, the Gemini LLM provider
• NATS messaging for event-driven transaction notifications
• FastAPI for web service implementation

Challenges we ran into

• Debugging complexity: When things went wrong with the Google ADK, debugging was challenging due to limited documentation. I had to read the open-source code directly to troubleshoot issues.
• ADK is bleeding edge: The version that I first used has bug in it, only after digging through Github issues I found out about the cryptic errors that I get and upgraded the version (the new version release was Sep 15, which is during the hackathon period!)
• Container registry requirements: Moving from local development (using kind) to GKE required hosting all container images in a registry, unlike local builds.

Accomplishments that we're proud of

• Seamless cloud deployment: Successfully transitioned from local development to GKE with minimal friction
• New tools never used before: ADK, A2A, Vertex AI and GKE are all new to me. But they are easy to get started with
• Agent-to-agent communication: Implemented AI agents that can communicate and delegate tasks to each other
• Real-time promotion checking system: Built an event-driven system that automatically detects and rewards qualifying transactions with AI agent as the logic
• Standardized AI tools: Created a robust MCP integration that makes banking actions accessible to AI agents

What we learned

• Observability needs: I learned that logging, traceability, and observability are essential for Kubernetes deployments, as debugging and monitoring become extremely difficult without them.
• Bleeding edge framework: Using bleeding edge framework is fun, but it can be a major blocker when something doesn't work as expected. Fortunately, the bug was fixed in a new release just one week before the submission.

What's next for Meet Your AI Bankers: Next-Gen Bank with AI Employees!

• Enhanced AI capabilities: Expand the range of banking services the AI agents can handle such as home loan agent
• Auth & Security: Authentication and session managements are very minimal
• Advanced promotion engine: Implement more sophisticated promotion logic and personalized offers
• Multi-channel support: Extend beyond chat to voice and mobile app integration
• Improved observability: Implement comprehensive monitoring and analytics dashboards
• Security enhancements: Add advanced fraud detection and security features
• Integration expansion: Connect with more external banking services and APIs

Built With

• a2a
• adk
• cronjob
• fastapi
• gke
• kind
• mcp
• nats
• python
• vertexai