AI Travel Planner

• 

  UI

• 

  Architecture Diagram

Inspiration

Travel planning is exciting but often overwhelming. Deciding which attractions to visit, where to eat, predicting traffic delays, and considering eco-friendly options can be tedious and time-consuming.

We were inspired to build AI Travel Planner after noticing that most travel apps provide static suggestions without personalization. We wanted a chat-based assistant that could plan trips intelligently, respond to user preferences, and even consider sustainability — all in one place.

What it does

AI Travel Planner is a chat-based AI assistant that:

• Generates personalized multi-day itineraries for any destination.
  
• Recommends restaurants, attractions, and activities.
  
• Predicts traffic delays and suggests optimal routes.
  
• Offers eco-friendly travel recommendations to reduce carbon footprint.
  
• Formats responses neatly in a ChatGPT-style UI.
  

Users simply type their trip preferences, and the AI produces a full, structured itinerary in response.

How we built it

1. Frontend:

   • Built with React and Tailwind CSS for a clean, chat-like UI.
     
   • Markdown rendering to display itineraries neatly.
     
   • Hosted on S3 + CloudFront for fast, secure delivery.
     

2. Backend:

   • AWS Lambda function in Python handles API requests.
     
   • Connected to API Gateway for RESTful access.
     
   • Lambda formats user input and calls Amazon Bedrock Nova LLM, parses responses, and sends them back to the frontend.

3. Workflow:

   • User sends request → Frontend → API Gateway → Lambda → Bedrock Nova → Lambda → Frontend → User sees formatted itinerary.

Challenges we ran into

• LLM input formatting: Amazon Bedrock Nova expects a precise messages array structure. Early attempts failed due to missing or malformed keys.
  
• Timeouts: Lambda initially timed out when LLM responses were large. Increased timeout and optimized input formatting solved this.
  
• CORS & deployment issues: Ensuring React + CloudFront + API Gateway communicated correctly required careful CORS header configuration.
  
• SPA routing: Refreshing React routes caused 404 errors; fixed by configuring CloudFront to serve /index.html for all routes.
  
• JSON parsing errors: Early responses from LLM sometimes returned unexpected text; added robust parsing and error handling.

Accomplishments that we're proud of

• Successfully integrated Amazon Bedrock Nova to generate structured travel itineraries from natural language input.
  
• Built a full-stack serverless app with React frontend, Lambda backend, API Gateway, and CloudFront hosting.
  
• Created a ChatGPT-like interface for user-friendly interaction.
  
• Added eco-friendly travel recommendations, a unique feature that enhances sustainability.
  
• Managed to deploy the app live and fully functional for public access.

What we learned

• How to integrate LLMs into a real-world application and handle structured input/output.
  
• Practical experience with serverless architecture using Lambda and API Gateway.
  
• Deploying a React frontend on S3 + CloudFront and solving SPA routing issues.
  
• Handling asynchronous responses, CORS, and JSON parsing reliably.
  
• The importance of formatting messages carefully for AI models and validating outputs.

What's next for AI Travel Planner

• Real-time traffic and weather integration to dynamically adjust itineraries.
  
• Local event recommendations to enhance user experiences.
  
• Booking links for restaurants and attractions.
  
• User profiles and saved trips for personalized suggestions.
  
• Voice interface for hands-free trip planning.
  
• Multi-modal support including images, maps, and visual itinerary cards.
  
• Eco-score and gamification to encourage sustainable travel choices.
  
• Mobile app version for on-the-go access.
  

These improvements will make AI Travel Planner more intelligent, personalized, and user-friendly, ensuring a seamless travel planning experience.

Built With

• amazon-web-services
• api
• bedrock
• nova
• python
• react