Building an Intelligent Banking Agent

Let's put it all together by building an intelligent banking agent system. This agent will handle multiple types of requests with specialized sub-agents:

Master Agent Architecture

The master agent coordinates between:

1. Verification Agent - Handles authentication
2. Loan Agent - Processes loan-related queries
3. Credit Card Agent - Manages credit card issues
4. Sentiment Analysis - Monitors customer satisfaction

Example code

 
class BankingState(TypedDict):
    customer_id: str
    query: str
    verified: bool
    sentiment: str
    response: str
    agent_type: str
 
def master_agent(state: BankingState) -> BankingState:
    """Coordinates between different specialist agents"""
    query = state["query"].lower()
    
    # Route to appropriate agent
    if "loan" in query:
        state["agent_type"] = "loan"
    elif "credit card" in query:
        state["agent_type"] = "credit"
    else:
        state["agent_type"] = "general"
    
    return state

What is the primary role of the master agent?

To analyze queries and route them to specialized agentsTo process all queries directlyTo store customer dataTo generate responses

Implementing Specialized Agents

Each specialized agent handles specific types of requests:

 
def verification_agent(state: BankingState) -> BankingState:
    """Verify customer identity"""
    # In production, this would check against database
    state["verified"] = True
    state["messages"].append("Identity verified successfully")
    return state
 
def loan_agent(state: BankingState) -> BankingState:
    """Handle loan-related queries"""
    if not state["verified"]:
        state["response"] = "Please verify your identity first"
        return state
    
    # Process loan query
    state["response"] = "Checking your loan payment status..."
    # Additional loan processing logic here
    return state
 
def sentiment_analyzer(state: BankingState) -> BankingState:
    """Analyze customer sentiment"""
    query = state["query"].lower()
    
    negative_words = ["frustrated", "angry", "disappointed"]
    if any(word in query for word in negative_words):
        state["sentiment"] = "negative"
        state["messages"].append("Escalating to specialist")
    else:
        state["sentiment"] = "neutral"
    
    return state

Continue

Complete Banking Agent Workflow

Let's assemble the complete banking agent system:

 
# Build the complete graph
banking_builder = StateGraph(BankingState)
 
# Add all nodes
banking_builder.add_node("master", master_agent)
banking_builder.add_node("verify", verification_agent)
banking_builder.add_node("loan", loan_agent)
banking_builder.add_node("credit", credit_card_agent)
banking_builder.add_node("sentiment", sentiment_analyzer)
 
# Add routing logic
def route_to_specialist(state: BankingState) -> str:
    if not state["verified"] and state["agent_type"] != "general":
        return "verify"
    return state["agent_type"]
 
# Connect the graph
banking_builder.add_edge(START, "sentiment")
banking_builder.add_edge("sentiment", "master")
banking_builder.add_conditional_edges(
    "master",
    route_to_specialist,
    {
        "verify": "verify",
        "loan": "loan",
        "credit": "credit",
        "general": END
    }
)
 
# Compile and ready to use
banking_graph = banking_builder.compile()

In this workflow, when is sentiment analysis performed?

At the beginning, before routing to the master agentAfter the specialized agent completesOnly for negative responsesNever - it's optional

Best Practices for Agentic AI Development

As you build agentic AI systems, follow these best practices:

1. State Management

• Keep state minimal and well-structured
• Use TypedDict for type safety
• Document state schema clearly

2. Error Handling

• Implement fallback paths for failures
• Log errors for debugging
• Gracefully handle edge cases

3. Testing

• Unit test individual nodes
• Integration test complete workflows
• Test conditional routing logic

4. Monitoring

• Track execution paths
• Monitor performance metrics
• Log decision rationale

Take the quiz