NoDetours: AI-Powered Travel Planning Assistant

NoDetours is a comprehensive travel planning application that uses Large Language Models (LLMs) to create personalized travel itineraries based on user preferences.

Application Overview

NoDetours creates detailed, personalized travel plans with:

• Custom Itineraries: Day-by-day schedules tailored to your preferences
• Packing Lists: Customized recommendations based on destination and activities
• Budget Estimates: Detailed cost breakdowns for different spending levels
• Calendar Integration: Export your itinerary directly to your calendar

The system processes natural language requests like "Help me plan a 7-day trip to Paris focusing on museums and local food" and generates comprehensive travel recommendations.

Key Features

• Natural Language Understanding: Simply describe your travel plans in plain English
• Preference Extraction: Automatically identifies destinations, durations, and preferences
• Contextual Information: Gathers weather forecasts, location details, and search results
• Multi-Modal Output: Provides itineraries, packing lists, and budget estimates
• Web Interface: Clean, intuitive UI with tabbed results display
• Calendar Export: Download your itinerary as an ICS file

Technical Architecture

The application consists of several components:

• Feature Extraction: Processes natural language input to identify travel preferences
• Search Query Generation: Creates effective search queries to gather relevant information
• Context Collection: Aggregates data from multiple sources (search, weather, maps)
• Output Generation: Produces detailed travel plans using an LLM
• Web Interface: Provides an intuitive user experience

Repository Structure

nodetours/
├── api/                          # API modules for external services
│   ├── app.py                    # FastAPI backend for web application
│   ├── llm_provider.py           # Unified interface for LLM providers
│   ├── maps.py                   # Maps API for location information
│   ├── scrape.py                 # Web scraping utilities
│   ├── search.py                 # Search API wrapper
│   └── weather.py                # Weather API for forecast data
├── app/                          # Core application modules
│   ├── agent.py                  # Main Travel Planner Agent
│   └── modules/                  # Specialized modules
│       ├── context_collector.py  # Information aggregation
│       ├── guardrail.py          # Input validation
│       ├── output_generator.py   # Travel plan generation
│       ├── search_query_extractor.py  # Feature extraction
│       └── search_query_generator.py  # Query generation
├── config/                       # Configuration files
│   ├── config.yaml               # Main configuration
│   └── eval_config.yaml          # Evaluation configuration
├── eval-data/                    # Evaluation datasets
│   ├── feature_extractor_data.json
│   ├── search_query_data.json
│   └── travel_assistant_data.json
├── evaluation_runs/              # Evaluation output directory
├── static/                       # Web UI assets
│   ├── css/
│   │   └── styles.css            # Application styling
│   └── js/
│       └── app.js                # Frontend JavaScript
├── templates/                    # HTML templates
│   └── index.html                # Main application page
├── utils/                        # Utility functions
│   └── helpers.py                # Helper utilities
├── evaluator.py                  # LLM evaluation module
├── generate_report.py            # Evaluation report generator
├── LICENSE                       # Apache 2.0 license
├── main.py                       # Application entry point
├── README.md                     # Project documentation
├── requirements.txt              # Python dependencies
└── run_evaluation.py             # Evaluation pipeline script

Installation and Setup

1. Clone the repository and set up your Python environment:

# Create and activate virtual environment (optional)
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

2. Create a .env file with your API keys:

OPENAI_API_KEY=your_openai_key
ANTHROPIC_API_KEY=your_anthropic_key
WEATHER_API_KEY=your_weather_api_key
MAPS_API_KEY=your_maps_api_key
FIRECRAWL_API_KEY=your_firecrawl_key

3. Run the application:

python main.py

For web interface:

python main.py --config config/config.yaml

For CLI mode:

python main.py --config config/config.yaml --cli

Usage

Web Interface

Once the application is running, access the web interface at http://localhost:8000 (or the configured host/port).

1. Enter your travel plans in the text input field, e.g., "Help me plan a 3-day trip to Chicago with museums, parks, and local food"
2. Click "Create Travel Plan" to generate your personalized itinerary
3. View your results in the tabbed interface:
   • Itinerary: Day-by-day travel plan
   • Packing List: Recommendations for what to bring
   • Budget: Estimated costs for your trip
4. Click "Download Itinerary Calendar" to export your plans to an ICS file

CLI Mode

When running in CLI mode:

1. Enter your travel plans when prompted
2. View your itinerary, packing list, and budget estimate in the console
3. Type 'exit' to quit

LLM Evaluation System

NoDetours includes a comprehensive evaluation system to compare different LLM providers for travel planning:

Evaluation Features

1. Tests multiple LLM providers (OpenAI GPT-3.5/4, Anthropic Claude, etc.)
2. Evaluates performance on various metrics (accuracy, relevance, completeness, usefulness, creativity)
3. Uses a "judge" LLM to rate responses
4. Generates detailed reports and visualizations

Running the Evaluation Pipeline

To run the full evaluation pipeline:

python run_evaluation.py --config config/eval_config.yaml --data eval-data/travel_assistant_data.json --sample-size 10

Parameters:

• --config: Path to the configuration file (default: config.yaml)
• --data: Path to the test data file (default: eval-data/feature_extractor_data.json)
• --sample-size: Number of test cases to sample (optional)
• --output-dir: Base directory for evaluation outputs (default: evaluation_runs)
• --skip-evaluation: Skip evaluation and use existing results file
• --results-file: Path to existing results file (if skipping evaluation)

Generating Reports from Existing Results

If you already have evaluation results and just want to generate reports:

python run_evaluation.py --skip-evaluation --results-file path/to/evaluation_results.json

Evaluation Metrics

The system evaluates travel plans on these metrics (scale 1-10):

• Accuracy: How accurately the plan addresses user requirements
• Relevance: How relevant the recommendations are to user preferences
• Completeness: How comprehensive and detailed the plan is
• Usefulness: How practical and helpful the information would be
• Creativity: How innovative and personalized the suggestions are

Configuration

You can modify the config.yaml file to:

1. Add or remove LLM providers
2. Change model parameters (temperature, max tokens)
3. Configure evaluation metrics
4. Set up API providers for weather, maps, and search

Example configuration for an LLM provider:

llm_providers:
  openai_gpt4:
    provider: "openai"
    model: "gpt-4"
    temperature: 0.7
    max_tokens: 4000

Extending the System

Adding a New LLM Provider

To add support for a new LLM provider:

1. Add the provider configuration to config.yaml
2. Ensure the LLMProvider class in api/llm_provider.py supports the new provider
3. Add the necessary API key to your .env file

Adding New API Integrations

To integrate a new external service:

1. Create a new wrapper class in the api/ directory
2. Update the config.yaml file with the new provider
3. Modify the context_collector.py to use the new service

Contributors

1. Surya Krishna Guttikonda
2. Monesh Rallapalli
3. Anish Sudarshan Gada
4. Prajwal Manohar

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.