Travel Chatter

An AI-powered travel companion for business travelers — plan trips, collaborate with teammates, discover things to do, and chat with an assistant that knows your trip's context.

Features

• Accounts & per-trip workspaces — register/login, create multiple trips, each with its own itinerary, wishlist, and chat history
• Per-trip AI chat — backed by Groq (Llama 3.3 70B). The assistant sees the trip's destination, dates, itinerary, wishlist, and recent messages
• AI can add itinerary items — tool/function calling; the assistant asks clarifying questions (date, time, airports, etc.) before adding anything
• Collaborative trips — invite other registered users by email. Collaborators can view the trip, chat with the AI, and submit itinerary requests that the owner approves/rejects
• Tinder-style discovery — on each trip, swipe through AI-generated activity suggestions. Swipe right = save to wishlist, swipe left = skip. Works with drag gestures or buttons
• 3D globe on My Trips — destinations appear as purple dots on a rotating night-earth globe (Globe.gl + Three.js). Clicking a trip card flies the globe to that location
• Geocoding — destinations are geocoded via OpenStreetMap's Nominatim (free, no API key)
• Rich itinerary — flights, hotels, meetings, activities, and other items with dates, times, locations, and notes

Tech Stack

• Backend: Node.js + Express, MongoDB (via Mongoose)
• Auth: bcrypt password hashing, JWT session tokens
• AI: Groq API with llama-3.3-70b-versatile (tool use enabled)
• Geocoding: OpenStreetMap Nominatim
• Frontend: Vanilla HTML/CSS/JS, dark-theme styling
• 3D Globe: Globe.gl (Three.js wrapper) loaded from unpkg

Setup

1. Install dependencies

   npm install

2. Create .env (copy from .env.example)

   MONGODB_URI=<your MongoDB connection string>
   JWT_SECRET=<long random string>
   GROQ_API_KEY=<your Groq key>
   PORT=3000
   

   • MongoDB: use Atlas (free tier) or install locally
   • Groq: free tier at console.groq.com
   • JWT_SECRET: generate with node -e "console.log(require('crypto').randomBytes(32).toString('hex'))"

3. Run

   npm start

   Open http://localhost:3000.

Pages

Path	Purpose
/index.html	Landing page
/login.html	Sign in
/register.html	Create account
/trips.html	My Trips — list + globe
/new-trip.html	Create a new trip
/trip-members.html?id=X	Manage people on a trip
/trip.html?id=X	Trip detail — itinerary, chat, wishlist, pending requests
/explore.html?id=X	Swipe-based activity discovery for a trip

How Chat Knows About the Trip

Every trip has its own chat history in MongoDB. On each user message, the server:

1. Loads the last 12 messages for that trip
2. Builds a system prompt with trip metadata (destination, dates, itinerary summary, wishlist, user role)
3. Calls Groq with an add_itinerary_item tool available
4. If the AI calls the tool, the server executes it (adding to itinerary for owners, creating a pending request for collaborators) and the AI confirms
5. Only the final text response is persisted

Security

• Passwords hashed with bcrypt
• JWT payload contains only email — no sensitive data
• helmet for standard security headers + Content Security Policy
• Rate limits: auth (10 / 15min per IP), chat (20/min per user), AI suggestions (10/hour per user), general API (60/min per user)
• Input length caps on all user-supplied fields to bound prompt-injection surface and DoS risk
• AI prompts explicitly mark trip context and user messages as untrusted data
• .env gitignored and never committed
• MongoDB connection uses TLS (ssl=true)

Data Model (MongoDB)

• User — email, passwordHash
• Trip — userId (owner), name, destination, lat/lng, startDate, endDate, purpose, status, itinerary[], wishlist[], members[], itineraryRequests[]
• ChatMessage — tripId, role (user/assistant), content, createdAt