• 

  icon_travelHK

About the Project: Travel Hong Kong

Inspiration

Travel planning in Hong Kong is often overwhelming due to the complexity of coordinating accommodation, transit, tickets, and attractions. Tourists also miss out on immersive cultural experiences and timely updates for popular sightseeing spots, leading to missed opportunities.

Travel Hong Kong aimed to create an AI-driven platform that personalizes travel itineraries based on individual interests and adapts dynamically to real-time weather conditions to help users enjoy a seamless and convenience experience.

What We Learned

Throughout this project, we gained valuable experience integrating AI with real-world data sources, such as live weather data from the Hong Kong Observatory. This experience also deepened my understanding of AI technologies such as embedding language model and voice synthesis. Additionally, our skills in designing interactive user interfaces that capture travelers' unique preferences is also be improved while we considered the user experience of the traveler. To implement advanced features like editable and savable itineraries in multiple formats, alongside voice-guided directions, we referenced many example to make the website contains more functions for the users.

How We Built It

Travel Hong Kong’s architecture is designed to balance intelligent backend processing with a responsive frontend interface. The core components include:

• User Input and Trip Setup: Users start by entering their origin location and trip duration. This foundational data frames the planning scope, ensuring the itinerary fits within the available timeframe.

• Weather Integration: Real-time weather information is automatically fetched from the Hong Kong Observatory API. This data guides the AI in recommending weather-appropriate activities, maximizing enjoyment and comfort.

• Preference Quiz: A tailored quiz solicits user interests in categories such as sightseeing, food hunting, cultural experiences, and unique landscapes. This preference model shapes the focus of the generated itinerary.

• AI-Generated Itinerary: Leveraging user inputs and weather data, the AI generates a detailed, day-by-day travel plan, specifying places to visit, dining options, and travel routes with estimated times for efficiency.

• Editable & Savable Plans: Users can customize their itinerary further and save it either to their cloud drive or export it as a PDF. Voice support is integrated for easy hands-free interaction during travel.

• AI Chat Room: To make this website more Hong Kong specific, we use Romer frog which is species unique in Hong Kong, as our mascot. An interactive chat interface allows users to provide additional information or ask questions to better tailor their travel plans, enabling dynamic communication with the AI model.

• Photo Guide: Users can upload images related to their planned visits, and the system provides descriptive information about the historical or cultural significance of the locations or objects in the photos, enriching their understanding of Hong Kong.

Challenges Faced

• ** Data Collection and Image Support :** Many parts of the app rely on images, such as photos of restaurants, hotels, and attractions, which enrich the user experience. However, obtaining free, high-quality images without copyright issues is difficult. To solve this, we integrated the Unsplash API, which provides access to a large library of free-to-use images. This allowed us to display attractive visuals legally and reliably.

• Integration of Multiple APIs Our project requires embedding various APIs to support different functionalities. For example, the real-time weather information is pulled from the Hong Kong Observatory’s API. Coordinating these external services and ensuring smooth communication with our website took significant time and development effort. Each API has its own authentication, rate limits, and data formats to manage properly.

• Language Model Embedding and Cost Control **The AI assistant at the core of Travel Hong Kong is based on advanced language models. We evaluated several options including OpenAI, Genmini, and others. Due to cost considerations, we ultimately chose to utilize the APIs available through **poe.com, balancing performance and budget. However, this solution still posed challenges like token limits, which constrain the complexity and length of user conversations we can handle without interruptions.

Built With

• amazonq
• css
• cursor
• hkoapi
• html5
• javascript
• node.js
• poe
• python