langgraph-travel-agent

English | 한국어

Production-ready LangGraph multi-agent system for travel planning. Async parallel tool orchestration across Amadeus, Hotelbeds, Twilio, and HubSpot.

What it does

• Natural-language travel request -> structured TravelPlan
• Parallel search across flights (Amadeus), hotels (Amadeus + Hotelbeds), activities (Amadeus)
• LLM-driven package generation (Budget / Balanced / Premium)
• Human-in-the-loop: customer-info form mid-conversation
• CRM (HubSpot) + SMS (Twilio) integration

Architecture

┌──────────────────────────────────────────────────────┐
│  FastAPI server (backend/api/main.py)                │
│    POST /chat -> task_id (async, background)         │
│    GET  /chat/status/{task_id}                       │
│    POST /chat/customer-info                          │
└────────────────────────┬─────────────────────────────┘
                         ▼
┌──────────────────────────────────────────────────────┐
│  LangGraph                                           │
│    call_model_and_tools  --collecting_info--> END    │
│    call_model_and_tools  --synthesizing----> synth   │
│    synthesize_results    -----------------> END      │
└────────────────────────┬─────────────────────────────┘
                         ▼
┌──────────────────────────────────────────────────────┐
│  Tools (parallel asyncio.gather)                     │
│    search_flights             Amadeus                │
│    search_and_compare_hotels  Amadeus + Hotelbeds    │
│    search_activities_by_city  Amadeus                │
│    send_sms_notification      Twilio                 │
│    send_to_hubspot            HubSpot                │
└──────────────────────────────────────────────────────┘

Layout

backend/
  api/main.py             FastAPI app
  config/settings.py      env loading, client init
  models/                 FlightOption, HotelOption, ActivityOption, TravelPackage, TravelPlan
  integrations/
    amadeus_client.py     hotel search, location conversion (airport / city / coords)
    hotelbeds_client.py   hotel search with X-Signature auth
  tools/
    flights.py            @tool search_flights
    hotels.py             @tool search_and_compare_hotels
    activities.py         @tool search_activities_by_city
    sms.py                @tool send_sms_notification
    crm.py                @tool send_to_hubspot
  graph/
    state.py              TravelAgentState
    analysis.py           enhanced_travel_analysis (request -> TravelPlan)
    nodes.py              call_model_node, synthesize_results_node, generate_travel_packages
    builder.py            wire nodes + conditional edges
  utils/helpers.py        offer parsing, time sorting, sampling, default dates
frontend/travel-widget/   React widget (separate package)
tests/                    pytest

Quick Start

git clone https://github.com/HarimxChoi/langgraph-travel-agent
cd langgraph-travel-agent
pip install -r requirements.txt

cp env.example .env
# fill GOOGLE_API_KEY, AMADEUS_API_KEY, AMADEUS_API_SECRET

uvicorn backend.api.main:app --host 0.0.0.0 --port 8000 --reload

POST a chat request:

curl -X POST http://localhost:8000/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "4-day trip from Seoul to Paris, $3000 budget", "thread_id": "demo-001"}'

Then poll /chat/status/{task_id} until status is completed.

Integration matrix

Service	Required	Used for
Google Gemini	yes	LLM (analysis, package generation, final response)
Amadeus	yes	Flights, hotels, activities
Hotelbeds	optional	Additional hotel inventory
Twilio	optional	SMS notifications
HubSpot	optional	CRM deal creation

Notes

• All external API calls run in parallel via asyncio.gather (e.g., flight + hotel + activity in one round-trip).
• Conformal logic uses Amadeus city codes for hotels and IATA codes for flights; LLM auto-converts between them.
• Customer info form is triggered on first turn (state: collecting_info); subsequent turns inject stored info.
• Replace HubSpot with another CRM by swapping the payload / endpoint in backend/tools/crm.py.

License

MIT