Travel Planner Assistant

An AI-powered travel planning assistant that helps users create personalized travel itineraries through conversation.

Click here to see demo

• Note: Due to my endsemester exams, I am able to just meet basic requirements of the project. I will be updating the project with more features and improvements in the future.

Requirements

• Python 3.9+
• SQLite3
• Ollama (for LLM)

Architecture Overview

Frontend (Streamlit)

• Built with app.py
• Uses Streamlit for the UI components and session management
• Features secure cookie-based authentication
• Real-time chat interface with message history
• Sidebar navigation for multiple chat sessions

Backend (FastAPI)

• Core API server in main.py
• JWT-based authentication system in auth.py
• SQLite database management in database.py
• AI chat logic in chatbot.py

Data Flow

1. Authentication Flow

   • User credentials → FastAPI /# or /register endpoints
   • JWT token generation and verification
   • Token stored in secure cookies and session state

2. Chat Flow

   • User message → Streamlit frontend
   • Request with auth token → FastAPI backend
   • Message processing with LLMs via Ollama:

     1. User Input → TinyLLAMA (Extracts key information from user prompt):

        • Location preferences
        • Date ranges
        • Budget constraints
        • Travel interests

     2. Processed Information → LLAMA2:

        • Generates detailed travel plans
        • Maintains conversation context
        • Provides recommendations
   • Response stored in database and returned to frontend

3. Database Structure

   • Users table: Authentication data
   • Chats table: Conversation sessions
   • Chat_messages table: Message history

AI Implementation

• Uses Ollama with LLaMA2 model
• Context-aware travel planning using TravelContext
• Progressive information gathering for:
  • Location
  • Dates
  • Budget
  • Interests
• Dynamic prompt engineering based on missing information

Setup Instructions

1. Install Dependencies

pip install -r requirements.txt

2. Setup Ollama (below for macOS):

• In one terminal, Install Ollama using Homebrew and serve the server:

brew install ollama
ollama serve

• In another terminal, pull the LLaMA2 and TinyLLaMA models:

ollama pull llama2
ollama pull tinyllama

3. Run FastAPI Server

uvicorn backend.main:app --reload

4. Run Streamlit App

streamlit run frontend/app.py

5. Open http://localhost:8501 in your browser and happy planning!

6. Checkout my demo here

Features

• Chat-based travel planning
• Persistent conversation history
• User authentication using Json Web Tokens
• Real-time AI responses
• Multi-chat support
• Weather preferences
• Budget constraints
• Location-based recommendations
• Date-based recommendations
• Interest-based recommendations
• Dynamic prompt generation

Future Work

The project can be extended in the following ways:

• Implement more advanced AI models
• Add support for multiple languages
• Integrate with external APIs for real-time data
• Map-based itinerary visualization with Map APIs
• Admin panel for user management

DataBase Schema

users (
    username TEXT PRIMARY KEY,
    password TEXT NOT NULL
)

chats (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    username TEXT NOT NULL,
    title TEXT NOT NULL,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
    FOREIGN KEY (username) REFERENCES users(username)
)

chat_messages (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    chat_id INTEGER NOT NULL,
    user_input TEXT NOT NULL,
    bot_response TEXT NOT NULL,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
    FOREIGN KEY (chat_id) REFERENCES chats(id) ON DELETE CASCADE
)