Travel-Path

• 

  plan

• 

  chat

Inspiration 💡

Group travel planning is often chaotic. While friends and family are full of suggestions, the actual planning—visualizing distances, comparing transport options, and structuring an itinerary—usually falls on one person. We were inspired by this common challenge. It's difficult for a group to collaboratively decide on a plan when they can't easily see how long it takes to get from one hotspot to another or compare the time and cost of a taxi versus public transport. Travel-Path was born from the need to make travel planning a visual, collaborative, and simple experience.

What It Does 🗺️

Travel-Path is an intelligent travel agent that generates detailed, optimized itineraries through a simple chat interface. Leveraging Hong Kong's open data, it provides real-time information to craft the perfect trip.

Key features include:

• Dynamic Itinerary Generation: Users can specify interests, and the app generates a logical travel plan.
• Data-Driven Insights: It integrates with data.gov.hk for accurate public transportation routes and timings, weather forecasts, and local events.
• Hyperlocal Discovery: It performs live web searches to identify the latest trending attractions, restaurants, and hidden gems for tourists.
• Visual Planning: The generated plan is displayed on an interactive map, giving users a clear visual reference of their journey.

How We Built It 💻

We developed a full-stack application using the following technologies:

• AI & Backend: We used Kiro to rapidly generate specifications and boilerplate code for our core functions. The backend logic is powered by the nova-lite LLM to process user requests and orchestrate the travel planning process.
• Frontend: We created an intuitive user interface featuring a chat window and an interactive map display.
• APIs & Data Sources: Our system is connected to a Geocoding API for location lookups, a Web Search API for real-time information discovery, and the rich datasets available through data.gov.hk.

Challenges We Ran Into 🚧

Our primary challenge was function integration and tool-use orchestration. While Kiro successfully generated individual functions for geocoding and web searches, wiring them into a cohesive system was complex.

Specifically, we encountered difficulties in prompt engineering. Our LLM, nova-lite, would sometimes default to its pre-trained knowledge instead of reliably calling our custom functions (like the geocoding or web search APIs). This meant that instead of fetching live, accurate data, it occasionally provided generic or "hallucinated" responses. Overcoming this is our key technical hurdle.

Accomplishments That We're Proud Of 🎉

Despite the challenges, we successfully built and deployed a functional full-stack application from scratch within the hackathon's timeframe. We are particularly proud of creating a seamless user experience that combines a conversational chat UI with a powerful, interactive map visualization—turning a complex planning task into a simple conversation.

What We Learned 🧠

This project underscored the critical importance of a modular and iterative development approach. We learned that generating individual AI functions is just the first step; creating a robust integration layer that ensures these functions are called reliably requires its own dedicated planning and rigorous testing. Moving forward, we'll break down integration into smaller, more manageable tasks to ensure system stability.

What's Next for Travel-Path 🚀

Our immediate priority is to refine the LLM's tool-use capability to ensure all integrations work flawlessly, transforming Travel-Path into a truly reliable AI travel agent.

Looking further ahead, our vision is to evolve Travel-Path into a platform that champions hyperlocal tourism. Many traditional and local businesses are closing down due to a lack of marketing resources. We plan to build a recommendation engine that actively promotes these smaller, authentic establishments, helping tourists discover the true culture of a city while supporting the local economy.

Built With

• api
• bedrock
• node.js
• react