Pet Store Agent - AWS Clash Competition

LaunchDarkly AI-powered agent for Virtual Pet Store customer service, built with LangGraph and deployed to AWS Bedrock AgentCore.

Quick Start

# Clone and setup
git clone <repo-url>
cd aws_clash
python3 -m venv venv
source venv/bin/activate
pip install -r agent/pet_store_agent/requirements.txt

# Configure environment
export LAUNCHDARKLY_SDK_KEY='your-sdk-key'
export AWS_PROFILE='your-profile'
export AWS_DEFAULT_REGION='us-east-1'

# Test locally
cd agent/pet_store_agent
python query_agent.py "What is the price of Doggy Delights?"

Repository Structure

aws_clash/
├── README.md                           # This file
├── agent/
│   ├── pet_store_agent/               # Main agent implementation
│   │   ├── README.md                  # Agent documentation & configuration
│   │   ├── pet_store_agent_full_ld.py # LaunchDarkly AI SDK integration
│   │   ├── tool_registry.py           # Tool definitions (5 tools)
│   │   ├── query_agent.py             # Local testing CLI
│   │   ├── storage/                   # Product catalog RAG index
│   │   └── data/                      # Source PDFs for RAG
│   └── knowledge_bases/
│       ├── README.md                  # PDF documentation
│       └── *.pdf                      # Product catalog PDFs
├── .env.example                       # Environment template
└── venv/                              # Python virtual environment

Key Features

✅ LaunchDarkly AI Configs - Dynamic model switching without redeployment ✅ LlamaIndex RAG - Product catalog retrieval from local PDFs ✅ LangGraph ReAct Agent - Tool orchestration with reasoning ✅ AWS Bedrock - Nova Pro & Claude models ✅ 5 Tools - 2 RAG + 3 Lambda/Mock ✅ Flexible Configuration - All params via LaunchDarkly custom config

Architecture

User Request
    ↓
LaunchDarkly AI Config (dynamic model selection)
    ↓
LangGraph ReAct Agent
    ├─→ retrieve_product_info (LlamaIndex RAG)
    ├─→ retrieve_pet_care (LlamaIndex RAG)
    ├─→ get_inventory (Lambda/Mock)
    ├─→ get_user_by_email (Lambda/Mock)
    └─→ get_user_by_id (Lambda/Mock)
    ↓
AWS Bedrock (Nova Pro / Claude)
    ↓
JSON Response

Configuration

All configuration managed through LaunchDarkly Agent API custom parameters:

{
  "aws_region": "us-west-2",
  "temperature": 0.7,
  "max_tokens": 4096,
  "use_real_lambda": true,
  "lambda_inventory_function": "team-PetStoreInventoryFunction-XXX",
  "lambda_user_function": "team-PetStoreUserFunction-XXX",
  "llamaindex_storage_dir": "./storage",
  "llamaindex_petcare_storage_dir": "./storage_petcare",
  "llamaindex_similarity_top_k": 5
}

See agent/pet_store_agent/README.md for complete parameter reference and configuration details.

LaunchDarkly Setup

1. Create AI Config at https://app.launchdarkly.com

   • Key: pet-store-agent
   • Provider: AWS Bedrock
   • Model: amazon.nova-pro-v1:0

2. Add Tools (in LaunchDarkly UI):

   • retrieve_product_info
   • retrieve_pet_care
   • get_inventory
   • get_user_by_email
   • get_user_by_id

3. Configure Custom Parameters (see above)

4. Add Instructions (system prompt)

5. Enable Targeting → Default rule → Select variation → Save

Deployment

Local Testing

cd agent/pet_store_agent
python query_agent.py "What is the price of Doggy Delights?"
python query_agent.py --interactive  # Interactive mode

AWS AgentCore Deployment

# Build Docker image for ARM64
docker build --platform linux/arm64 -t petstore-agent:latest \
  -f agent/Dockerfile agent/

# Push to ECR
aws ecr get-login-password --region us-east-1 | \
  docker login --username AWS --password-stdin <account>.dkr.ecr.us-east-1.amazonaws.com
docker tag petstore-agent:latest <account>.dkr.ecr.us-east-1.amazonaws.com/petstore-agent:latest
docker push <account>.dkr.ecr.us-east-1.amazonaws.com/petstore-agent:latest

# Deploy to AgentCore
cd agent
python deploy_to_agentcore.py

Competition Strategy

Key Advantage: Update configuration in LaunchDarkly without redeployment!

Submit ARN → Get Score → Update LaunchDarkly Config → Retest → Resubmit

No container rebuild or redeployment needed - iterate on:

• Model selection (Nova Pro / Claude Sonnet / Claude Haiku)
• Temperature and max_tokens
• System prompt/instructions
• Tool parameters

Testing

Local:

cd agent/pet_store_agent
python query_agent.py "query here"

Interactive:

python query_agent.py --interactive

Deployed:

aws bedrock-agentcore invoke-agent-runtime \
  --agent-runtime-arn "arn:aws:bedrock-agentcore:us-east-1:<account>:runtime/<id>" \
  --qualifier DEFAULT \
  --content-type application/json \
  --payload '{"prompt":"What is the price of Doggy Delights?"}' \
  --region us-east-1 \
  response.json

Documentation

• agent/pet_store_agent/README.md - Agent implementation, configuration, and parameters
• agent/knowledge_bases/README.md - PDF documentation
• .env.example - Environment variable template

Monitoring

• CloudWatch Logs: aws logs tail agents/petstore-agent-logs --follow
• LaunchDarkly Dashboard: AI Config metrics, token usage, errors
• AWS Console: AgentCore runtime status and health

Troubleshooting

Agent not responding?

aws bedrock-agentcore-control get-agent-runtime \
  --agent-runtime-id <runtime-id> --region us-east-1

LaunchDarkly connection failed?

• Check SDK key in environment variables
• Verify CloudWatch logs for initialization errors
• Ensure network mode is PUBLIC in runtime config

Import errors?

pip install -r agent/pet_store_agent/requirements.txt

Built With

• LaunchDarkly AI SDK - Dynamic AI configuration
• LangGraph - ReAct agent framework
• LlamaIndex - Local RAG implementation
• AWS Bedrock - Nova Pro & Claude models
• AWS AgentCore - Containerized runtime

---

For detailed implementation and configuration: agent/pet_store_agent/README.md