Building a Proactive AI Travel Agent on AWS: My Journey with Bedrock AgentCore (Part 2)

🧠 From One Agent to a Multi-Agent System

In Part 1, I walked through the foundation: getting a single AI agent running on Amazon Bedrock AgentCore. That first step gave me a working runtime, a clean HTTP interface, and confidence that I could build a travel concierge on top of it.

But a real travel concierge needs more than just chat. It needs to:

Understand complex travel requests
Plan multi-step itineraries
Search for flights and hotels
Manage bookings
Be modular enough to extend and scale

That’s exactly what Phase 2 is all about.

🏗️ Phase 2: Local Multi-Agent Travel System

In this phase, I focused on building a fully functional travel orchestration system — still local, but structured to integrate cleanly with Bedrock AgentCore.

💻 GitHub repo: bedrock-travel-agent

🎯 Goals

Modular architecture separating reasoning, planning, and execution
A realistic set of travel tools (flights, hotels, bookings) for testing
Clean interfaces that map easily to Bedrock AgentCore’s tool calling
Keep it Bedrock-ready — no rewriting later

🛠️ Core Components

File	Role
`orchestrator.py`	The “brain” — routes user requests to the right agents and tools
`planning_agent.py`	Interprets natural language, extracts intent, decides actions
`travel_tool.py`	Implements flight and hotel search (ready for real APIs)
`booking_tools.py`	Manages bookings: create, confirm, cancel
`planning_tools.py`	Handles budgets, date parsing, itinerary generation
`mock_travel_api.py`	Provides realistic test data locally
`my_agent.py` / `run_agent.py`	Future Bedrock AgentCore integration entry points

🔄 How It Works

User Input → Orchestrator → Planning Agent → Travel Tools → Response

User: "Plan a 5-day trip to Paris"
↓
Planning Agent: Understands intent, breaks it into sub-tasks
↓
Travel Tools: Fetch flights, hotels, estimate budget
↓
Response: Detailed itinerary and costs

🚦 Sample Interactions

User: “Find flights from NYC to London”

Agent: Returns flights with prices and timings.

User: “I need hotels in Tokyo for next week”

Agent: Returns hotels with ratings, prices, and amenities.

User: “Plan a 5-day trip to Paris”

Agent: Suggests day-by-day itinerary, estimated costs, and booking options.

🧩 Architecture Overview

🏁 Current Status: Ready for Bedrock

At this point, I’ve got:

✅ A multi-agent travel orchestration system

✅ Clean separation of logic

✅ Tools and agents that match Bedrock’s tool-calling paradigm

✅ Mock APIs for safe, realistic testing

But everything is still local. AWS doesn’t know about it — yet.

🚀 What’s Next (Phase 3)

In the upcoming phase, I’ll:

Define tool schemas for Bedrock AgentCore
Register these tools and agents with AgentCore
Deploy the entire system into AWS for scaling, routing, and observability
Test real Bedrock-driven multi-agent orchestration in production-like conditions

Please drop you comments in you have any questions while performing the above.

Thank you,
Harsha