Travel Assistant

Interactive AI travel planner

🚀 AI-Powered Travel Assistant

Multi-agent orchestration with LangGraph • Azure LLMs • React • FastAPI • MongoDB

🐱 Repository · 📍 Live Demo

An AI‑powered travel planning app. Users chat with an assistant to design multi‑city trips, receive a day‑by‑day itinerary, hotel suggestions, and an inter‑city transport plan. Backend: FastAPI + MongoDB. Frontend: React + Material UI.

Highlights

• Multi‑agent itinerary generation (destination selection, routing, detection/modification, itinerary writer)
• Transport plan with realistic durations and heuristic costs between cities
• Hotel suggestions generated in background after itinerary changes
• Real‑time chat over WebSocket with streaming assistant messages
• JWT authentication and protected routes (bypassed in demo)
• Modern React + MUI interface with tabs: Chat, Itinerary, Hotels, Transport

Feature Deep‑Dive (What each tab really does)

Walkthrough

1) Login (mock bypassed in demo)

2) Trip configuration (hidden in demo, auto‑seeded)

3) Chat with the AI assistant

Powered by a GPT‑class Large Language Model (LLM) that performs intent detection and conversation orchestration.
User requests (e.g., "cultural trip", "add beach days at the end") trigger the workflow to either create a fresh plan or modify the existing one.
The assistant streams responses through WebSocket and persists messages linked to the current travel_id.

4) Itinerary: dates and cities

The itinerary is produced by a LangGraph orchestrated pipeline coordinated by SmartItineraryWorkflow:
1) A Routing Agent builds a graph where nodes are cities (loaded with their coordinates from MongoDB) and edges represent geodesic distances.
2) The agent proposes a near‑optimal visiting order (shortest‑path / TSP‑like heuristic) across all requested cities.
3) An Itinerary Agent (LLM) writes the day‑by‑day plan aligning with the route and user preferences.
4) A Time Budget Scheduler assigns hours per city and feeds downstream services (hotels and transport).
In the UI you see each city with explicit date ranges (arrival → departure) derived from this schedule.

5) Hotels: suggestions per city window

A Hotel Suggestions Service calls an external provider API using the itinerary windows (city + check‑in/check‑out) and the user's preferences.
When the itinerary is changed from the chat (e.g., cities reordered or days shifted), the hotel suggestions are regenerated in the background and the UI reflects the new results.
In the demo (mock) build, these suggestions are deterministic and consistent with the displayed itinerary.

6) Transport: inter‑city plan

The transport plan is currently heuristic‑based but fully respects the itinerary's city sequence and durations.
It selects a mode per leg (car/bus/train/flight/boat) using distance thresholds and adds realistic overheads for total travel time; costs are computed with simple per‑km + fixed‑fee models.
Future work can plug real providers without changing the frontend.

---

Quick Start (Mocked Demo)

The demo runs entirely on the frontend with predefined data. No backend or database required.

1. From the repo root:
   • cd frontend
   • npm install
   • REACT_APP_MOCK=true npm start
2. The app opens at http://localhost:3000 and skips login, going straight to the chat.

• Frontend: set REACT_APP_MOCK=true to enable demo mocks. The app seeds localStorage (token/user) and bypasses login.
• Backend: set MOCK_MODE=true to return predefined responses (optional; not needed for the GitHub Pages demo).

Deployment to GitHub Pages (gh-pages workflow) is already configured in .github/workflows/gh-pages.yml.

---

Architecture (One‑Page Overview)

Backend (FastAPI + MongoDB)

• Routers under /api, CORS/security middlewares, async Mongo (Motor)
• Multi‑agent orchestration with LangGraph:
  • SmartItineraryWorkflow: coordinates itinerary detection/modification, DB reads, routing, itinerary creation
  • RoutingAgent: builds a graph and proposes the route order (distance heuristics / TSP‑like)
  • ItineraryAgent: generates day‑by‑day plans from the route + preferences
  • TimeBudgetScheduler: allocates hours per city using total days and transport matrix
  • TransportPlanService: creates the inter‑city plan after itinerary changes
• Background jobs: hotel suggestions + transport plan generation after itinerary updates
• WebSocket: /api/travels/{travel_id}/ws for live chat

Frontend (React + MUI)

• MainCanvas layout with tabs: Chat, Itinerary, Hotels, Transport
• In demo mode the app intercepts fetch and WebSocket to serve mock data
• HashRouter + homepage support for GitHub Pages

---

Local Development (Full Stack)

Prerequisites

• Python 3.10+
• Node.js 18+
• MongoDB (local or remote)
• Azure OpenAI credentials (if not using mock)

Backend

1. cd backend
2. pip install -r requirements.txt
3. Create .env with at least:
   • HOST=0.0.0.0, PORT=8000, SECRET_KEY=change_me
   • MONGODB_URL=mongodb://localhost:27017, DATABASE_NAME=travel_app
   • AZURE_OPENAI_* variables (omit if MOCK_MODE=true)
   • MOCK_MODE=false (or true for a backend demo)
4. Start: uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

Frontend

1. cd frontend
2. npm install
3. npm start (or REACT_APP_MOCK=true npm start for demo‑only)

---

GitHub Pages (Live Demo)

This repository uses a gh-pages branch deployment:

• Workflow: .github/workflows/gh-pages.yml
• Build command: npm ci && REACT_APP_MOCK=true npm run build (in frontend/)
• Output: frontend/build → published to gh-pages
• SPA fallback: 404.html copies index.html

Public link: https://amarkosmarkos.github.io/TravelApp/

---

API Overview (Backend)

All endpoints are under /api (JWT‑based auth in routers/auth.py). Highlights:

• Travels: GET/POST /api/travels, GET/PUT/DELETE /api/travels/{id}
• Itinerary: GET/POST /api/travels/{id}/itinerary
• Hotels: GET /api/travels/{id}/hotels/suggestions
• Transport: GET /api/travels/{id}/transport-plan
• WebSocket: GET /api/travels/{id}/ws?token=...

After itinerary changes, background tasks regenerate hotels and transport automatically.

---

License

This work is licensed under the Creative Commons Attribution‑NonCommercial 4.0 International License (CC BY‑NC 4.0).

© 2024 All rights reserved.