AI Trip Planner

An AI-powered travel assistant that plans end-to-end trips (flights, hotels, car rentals) based on user preferences, with multilingual support and rich UI rendering on the web.

Repository Structure (brief)

AI-Trip-Planner/
  backend/
    app/
      __init__.py
      main.py           # FastAPI app entry
      agents.py         # Guardian Agent, TravelAgent wiring
      travelAgent.py    # TravelAgent logic and tool usage
    models/
      Activity.py
      Chat.py
      Message.py
  frontend/
    src/
      App.tsx
      components/
        AIChatbot.tsx
        ItineraryBuilder.tsx
        figma/ImageWithFallback.tsx  # hotel image fallback
        ui/ ... (shadcn/ui components)
      styles/globals.css
      main.tsx
    index.html
    vite.config.ts

Main Logic (high level)

• User converses in the web UI (AIChatbot.tsx).
• Backend (FastAPI) orchestrates an LLM-based flow to:
  • understand intent and preferences (dates, budget, cabin class, chains, car class, etc.),
  • fetch flight/hotel/car options via connected tools,
  • evaluate and rank options against user preferences,
  • build an itinerary and stream results back to the UI.

LLM Architecture

• Guardian Agent → TravelAgent → Tools

Flow:

1. Guardian Agent

   • Validates and guards prompts, enforces policies and safety.
   • Sets constraints and high-level objectives for planning.

2. TravelAgent (planner/executor)

   • Decomposes the task (transport → stay → ground).
   • Calls domain tools, merges results, and produces the trip plan.

3. Connected Tools (used by TravelAgent)

   • MCP: external data access (e.g., flight/hotel/car providers) and utilities.
   • pick_options: scores and re-ranks candidate options based on user preferences.
   • trip_plan: composes the final, day-by-day plan and summaries.

Guardian Agent → TravelAgent → [ MCP | pick_options | trip_plan ] → Ranked options + Itinerary

Preference-aware Selection

• Flights: balances price, departure/arrival windows, cabin class, baggage, airline preference, and number of stops (prefers direct when requested).
• Hotels: considers price/night, location proximity, rating, amenities, cancellation policy, and brand/chain preferences; images are rendered in UI with a fallback.
• Car rentals: weighs price, pickup/dropoff alignment, transmission/class preference, mileage/insurance rules, and provider reliability.
• Re-ranking is adaptive: the system re-scores results when the user changes constraints (e.g., “evening flight”, “walkable to city center”).

Language Support

• Multilingual conversation handling: prompts and responses support multiple languages; Turkish and English are first-class.
• The UI and agent respond in the user’s language when possible.

Hotel Image Rendering

• The frontend uses components/figma/ImageWithFallback.tsx to display hotel images.
• Graceful fallback avoids layout shifts or broken images, improving UX for listings and plan cards.

Frontend Highlights

• React + Vite + TypeScript.
• Components: AIChatbot for conversation, ItineraryBuilder for structured trip output.
• Reusable ui/ components (shadcn/ui) for a modern, consistent interface.

Backend Highlights

• FastAPI application (app/main.py).
• Agents defined in app/agents.py; planning logic in app/travelAgent.py.
• Pydantic models in backend/models for chats, messages, and activities.

Notes

• This README focuses on architecture and capabilities. Internal endpoints, auth, and provider credentials are expected to be configured per environment.