AgentVerse: AI-Driven Customer Engagement

AgentVerse: A universe of AI agents working together to revolutionize customer engagement

Comment

Inspiration

Customers today demand instant and personalized service. Traditional customer support systems often struggle to keep up, leading to long wait times and frustrating experiences. We were inspired to create a solution that leverages the power of AI to provide seamless and efficient customer engagement, 24/7. EazyBank, a fictional but forward-thinking bank, provided the perfect context to explore this vision. We wanted to demonstrate how AI agents, working in concert, can revolutionize how banks interact with their customers.

What it does

AgentVerse is a multi-agent AI application designed to transform customer engagement for EazyBank. It allows customers to interact with a universe of specialized AI agents to get instant answers to their queries. AgentVerse can:

  • Provide real-time account status updates based on registered phone number.
  • Explain the reasons for account application rejections, drawing detailed information from bank policy documents.
  • Offer insights into EazyBank's market performance, including stock prices and relevant news.
  • Seamlessly transition complex or sensitive requests to a human agent, preserving context and conversation history.

How we built it

We built AgentVerse using Google Cloud's Agent Development Kit (ADK) to orchestrate the interaction between four specialized AI agents.

  • Agent 1 (Account Status Agent): We created an OpenAPI specification for calling a Cloud Run service that fetches account application statuses. This agent uses the user's identifier (phone number) to query the API.
  • Agent 2 (Rejection Reason Agent): We implemented a RAG (Retrieval-Augmented Generation) system using Vertex AI Search and Conversation. This agent accesses a Datastore containing PDF documents with detailed rejection reasons, enabling it to provide clear and contextual explanations.
  • Agent 3 (Market Insights Agent): We leveraged the built-in Google Search tool within ADK to enable this agent to fetch the latest EazyBank stock prices and market news.
  • Agent 4 (Human Handoff Agent): This agent detects when a human agent is needed. It packages the conversation history and user data and publishes a message to a Pub/Sub topic. A separate customer service platform subscribes to this topic and assigns the case to a human agent.

We used Python for the agent logic and Cloud Run Functions to deploy Agent 1. IAM permissions were configured to allow the Cloud Run Function to publish to the Pub/Sub topic.

Challenges we ran into

  • Agent Orchestration Complexity: Designing the flow of information and transitions between agents proved challenging, especially ensuring that the context was properly preserved.
  • RAG Implementation: Setting up the RAG system with Vertex AI Search and Conversation required significant effort to index the PDF documents in the Datastore and optimize the query process.
  • Human Handoff Integration: Connecting the Pub/Sub topic to a realistic customer service platform was beyond the scope of the hackathon, so we focused on demonstrating the Pub/Sub publishing mechanism.
  • OpenAPI Specification: Defining the OpenAPI specification for Agent 1 to call the Cloud Run service required careful attention to detail to ensure proper data exchange.

Accomplishments that we're proud of

  • Functional Multi-Agent System: We successfully built a working multi-agent system that demonstrates the power of ADK for customer engagement.
  • Seamless Agent Transitions: We achieved smooth transitions between agents, preserving the context of the conversation.
  • RAG-Powered Rejection Explanations: We were able to provide detailed and contextual explanations for account application rejections using the RAG system.
  • Human Handoff Mechanism: We implemented a robust mechanism for handing off complex requests to human agents.
  • End-to-End Demonstration: We created a compelling end-to-end demonstration of how AgentVerse can transform customer engagement for EazyBank.

What we learned

  • The Power of ADK: We learned how ADK can be used to create sophisticated multi-agent systems with relative ease.
  • Importance of Context: We realized the crucial role of preserving context during agent transitions to provide a seamless user experience.
  • Challenges of RAG Implementation: We gained hands-on experience with the complexities of setting up and optimizing a RAG system.
  • Value of OpenAPI Specifications: We understood the importance of well-defined OpenAPI specifications for agent communication and API integration.
  • Cloud Pub/Sub for Event-Driven Architectures: We learned how Cloud Pub/Sub can be used to build event-driven architectures for seamless integration between different systems.

What's next for AgentVerse: AI-Driven Customer Engagement

  • Full Integration with a Customer Service Platform: We plan to fully integrate AgentVerse with a real-world customer service platform to provide a complete human handoff experience.
  • Advanced Agent Training: We will explore advanced training techniques to improve the accuracy and capabilities of the AI agents.
  • Personalized Recommendations: We aim to incorporate personalized recommendations based on user history and preferences.
  • Proactive Customer Engagement: We envision AgentVerse proactively engaging with customers to provide timely updates and assistance.
  • Expand Agent Capabilities: We plan to add more specialized agents to handle a wider range of customer inquiries and tasks, such as fraud detection, loan applications, and financial planning.
  • Enhance RAG System: We will continue to refine the RAG system to improve the accuracy and relevance of the rejection explanations.

Built With

Share this project:

Updates

Submission history