AA Agentic AI Workshop

Build and deploy an AI-powered Flight Operations agent using Strands SDK and Amazon Bedrock AgentCore — driven by Kiro's spec-driven development workflow.

The Scenario

A passenger's connecting flight through Dallas is delayed. They ask:

"My connection in Dallas is delayed. What are my rebooking options to get to Chicago?"

Your agent handles this in a single turn — checking flight status, traversing the route knowledge graph to find alternatives avoiding DFW, and presenting rebooking options with seat availability.

What You'll Learn

• Build agents with Strands SDK custom @tool functions
• Use a knowledge graph (NetworkX) for multi-hop route reasoning
• Run locally with the dual-mode entrypoint pattern (FastAPI ↔ AgentCore Runtime)
• Deploy to AgentCore Runtime via CDK
• Govern tools with AgentCore Gateway access policies
• Monitor with AgentCore Observability (traces, spans, metrics)

Quick Start

# Install dependencies
pip install -e .

# Verify AWS access (Nova Pro model required)
aws sts get-caller-identity
aws bedrock list-foundation-models --query "modelSummaries[?contains(modelId, 'nova')]"

# Run locally
RUNTIME_MODE=local python -m src.entrypoint_local

# Test with CLI
python cli.py

Workshop Structure

Module	Duration	Activity
1. Kiro Setup	15 min	Install Powers, dependencies, verify AWS
2. Spec Planning	30 min	Review requirements → design → tasks, generate scaffolding
3. Build Agent	45 min	Write tools, graph client, agent definition
4. Test Locally	20 min	Run agent, test Dallas disruption scenario
5. Run Tests	15 min	pytest validation
6. Deploy to AWS	30 min	CDK deploy to AgentCore Runtime
7. Explore Gateway	20 min	Tool registrations, access policies
8. Observability	15 min	CloudWatch traces and spans
9. Rebooking Agent	40 min	(Optional) Multi-agent with distinct permissions
10. Neptune Extension	30 min	(Optional) Local graph → Amazon Neptune

Total: ~3 hours core | ~4 hours with extensions

Project Structure

aa-workshop-agent/
├── .kiro/
│   ├── specs/aa-workshop-agent/    # Requirements, design, tasks
│   └── steering/aa-workshop.md     # Project conventions for Kiro
├── src/
│   ├── agent.py                    # Strands Agent (Nova Pro)
│   ├── tools.py                    # @tool functions
│   ├── graph_client.py             # NetworkX / Neptune abstraction
│   ├── data_store.py               # JSON data loader
│   ├── entrypoint_local.py         # FastAPI wrapper
│   └── entrypoint_aws.py           # AgentCore Runtime wrapper
├── data/
│   ├── flights.json                # Flight, seat, passenger data
│   └── routes.json                 # Airport graph (nodes + edges)
├── deployment/
│   └── cdk_app.py                  # CDK stack
├── tests/                          # Unit + property-based tests
├── cli.py                          # Interactive CLI
├── WORKSHOP_GUIDE.md               # Full step-by-step instructions
└── docs/workshop-site/workshop.html # Standalone branded microsite

Prerequisites

• Python 3.11+
• Node.js 20+ (CDK CLI)
• AWS CLI configured with valid credentials
• Amazon Bedrock access for Nova Pro model
• Kiro IDE with Powers: aws-agentcore, strands, cloud-architect

Resources

• Workshop Guide — Full step-by-step instructions
• Microsite — Open in browser (standalone, no server needed)
• Strands SDK Docs
• AgentCore Docs

Architecture

Same agent code → different infrastructure

Local:  agent.py → FastAPI/uvicorn → NetworkX graph → JSON files
Cloud:  agent.py → AgentCore Runtime → Neptune (optional) → Gateway governance
                                                           → CloudWatch traces

The dual-mode pattern separates agent logic from transport and data layers. Your code never changes between environments.