Skip to content

README.md

Latest commit

 

History

History
153 lines (114 loc) · 6.42 KB

README.md

File metadata and controls

153 lines (114 loc) · 6.42 KB

Customer Service

AI-powered customer service resolution for banking, combining inquiry handling, transaction investigation, and product advisory to deliver comprehensive support in a single interaction.

Overview

The Customer Service application routes incoming banking inquiries to specialized agents that classify issues, investigate transaction disputes, and recommend products. The orchestrator synthesizes agent outputs into a unified resolution with clear next steps, priority levels, and follow-up actions.

Business Value

  • Faster Resolution -- Parallel agent execution reduces average handling time for complex inquiries
  • Consistent Quality -- Standardized workflows ensure every inquiry follows bank policies and procedures
  • Cross-Sell Opportunities -- Product advisory agent surfaces relevant product recommendations during service interactions
  • Reduced Escalations -- AI-driven dispute investigation and FAQ handling resolves more issues at first contact
  • Scalable Support -- Handle peak volumes without proportional staffing increases

Architecture

graph TB
    Request["Client Request<br/><small>Customer Inquiry</small>"] --> Runtime["AgentCore Runtime"]
    Runtime --> Orchestrator["Orchestrator"]
    
    Orchestrator --> Router{{"Inquiry<br/>Handler"}}
    
    Router --> TransactionAgent["Transaction Specialist<br/><small>Dispute resolution & payment issues</small>"]
    Router --> ProductAgent["Product Advisor<br/><small>Product recommendations & guidance</small>"]
    Router --> FAQ["FAQ Response"]
    
    TransactionAgent --> Bedrock["Amazon Bedrock<br/>(Claude)"]
    ProductAgent --> Bedrock
    
    TransactionAgent --> S3["S3 Sample Data<br/><small>data/samples/customer_service/</small>"]
    ProductAgent --> S3
    
    TransactionAgent --> Response["Response<br/><small>Resolution & recommendations</small>"]
    ProductAgent --> Response
    FAQ --> Response
    
    style Request fill:#e1f5ff
    style Response fill:#d4edda
    style Bedrock fill:#fff3cd
    style S3 fill:#fff3cd
    style Router fill:#f8d7da
Loading

Directory Structure

use_cases/customer_service/
├── README.md
└── src/
    ├── __init__.py                              # Framework router
    ├── strands/
    │   ├── __init__.py
    │   ├── config.py                            # Service settings
    │   ├── models.py                            # ServiceRequest / ServiceResponse
    │   ├── orchestrator.py                      # CustomerServiceOrchestrator
    │   └── agents/
    │       ├── inquiry_handler.py               # InquiryHandler agent
    │       ├── transaction_specialist.py        # TransactionSpecialist agent
    │       └── product_advisor.py               # ProductAdvisor agent
    └── langchain_langgraph/                     # LangGraph implementation (same structure)

Agentic Design

The CustomerServiceOrchestrator extends StrandsOrchestrator and implements a routing + parallel pattern:

  1. Routing -- The inquiry_type field determines which agents are invoked: full runs all three in parallel; general, transaction_dispute, and product_inquiry route to a single specialist; service_request runs the Inquiry Handler and Transaction Specialist in parallel.
  2. Parallel Execution -- For full assessments, all three agents run concurrently via asyncio.gather().
  3. Synthesis -- A supervisor LLM call produces a resolution status (RESOLVED/PENDING/ESCALATED), priority level, actions taken, follow-up requirements, and product recommendations.

Agents

Inquiry Handler

Field Detail
Class InquiryHandler(StrandsAgent)
Role Classifies customer inquiries by type and urgency, routes to specialists, handles FAQs
Data Customer profile, service history via s3_retriever_tool
Produces Inquiry classification, priority level (LOW/MEDIUM/HIGH/URGENT), routing recommendation, estimated resolution timeframe

Transaction Specialist

Field Detail
Class TransactionSpecialist(StrandsAgent)
Role Investigates transaction disputes, assesses refund eligibility, resolves payment issues
Data Customer profile, transaction history via s3_retriever_tool
Produces Transaction analysis, dispute classification, refund eligibility, resolution recommendation

Product Advisor

Field Detail
Class ProductAdvisor(StrandsAgent)
Role Analyzes customer profiles and recommends suitable banking products
Data Customer profile, product catalog via s3_retriever_tool
Produces Customer needs analysis, ranked product recommendations, feature comparisons, personalized guidance

Data and Tools

  • Tool: s3_retriever_tool -- Retrieves customer data from S3 by customer ID and data type
  • S3 Path: data/samples/customer_service/{customer_id}/
  • Data Files: profile.json (account info, products, history)

Request / Response

Request (ServiceRequest)

class ServiceRequest(BaseModel):
    customer_id: str                               # e.g. "CUST001"
    inquiry_type: InquiryType = "full"             # full | general | transaction_dispute | product_inquiry | service_request
    additional_context: str | None = None

Response (ServiceResponse)

class ServiceResponse(BaseModel):
    customer_id: str
    service_id: str                                # UUID
    timestamp: datetime
    resolution: ResolutionDetail | None            # status, priority, actions_taken, follow_up_required, notes
    recommendations: list[str]                     # Product/service recommendations
    summary: str                                   # Executive summary
    raw_analysis: dict

Quick Start

# Deploy to AgentCore
USE_CASE_ID=customer_service ./scripts/deploy/full/deploy_agentcore.sh

# Test
./scripts/use_cases/customer_service/test/test_agentcore.sh

Sample Data

Customer ID Profile Description
CUST001 Premium Checking Established customer with multiple products

Related Documentation