GlobalMate - AI Assistance for Travel Guide

Are you tired of using Google and trying to find flight details, hotel information and planning restaurants for the trip? Here is our AI Agent, GlobalMate, to help make these tasks easy for you.

Comment

Inspiration

We were discussing what ideas we can use to make an Agent and came up with the idea of using a health service agent, but while discussing details, we realised a key problem. The key problem is that, as it is an AI, the user may not feel safe sharing their private details about their health or think it might give highly inaccurate feedback. This will be a major issue, as if they believe this and any user encounters wrong information about their health, they will instantly label the agent dangerous. So we took a step back and started to think we all things can be considered a problem in day-to-day life. After a few days, we still could not come up with the idea and were near the idea of unregistering from the hackathon, but one of our friends started asking us if we knew any place to find information about flights, hotel information and restaurants to eat at. As our friend was planning to take his group of friends abroad for his club trip. It took us some time, but we eventually found the information needed and sent him and also realised why not make an agent which could do all of this work for us and does not rely on other human ideas, which may contain biases. This was our inspiration for the project.

What it does

As the name suggests, it is a travel guide which provides:

  • Flights – Find and book flights with real-time AI assistance
  • Hotels – Discover and reserve your perfect stay
  • Restaurants – Find and book tables with AI suggestions
  • Weather – Get live weather updates for your destination
  • Price Alerts – Stay informed with Fetch.ai-powered hotel price tracking
  • Chat Interface – Powered by Gemini Pro 1.5 with voice input and read-aloud support

How we built it

We have used the gemini api as a base for our chatbot, then added the function to search Google for questions outside the set predefined responses. At its core, GlobalMate employs a multi-layered architecture designed for both responsiveness and scalability. The backend is built on Python's Flask framework, chosen for its lightweight nature and flexibility in handling our diverse API integrations. We implemented a modular design pattern that separates concerns into distinct components: natural language understanding, data aggregation, and user interface rendering. This separation proved crucial when we needed to swap out the Zomato API for Yelp's platform mid-development due to changing API availability. Our data pipeline connects to three major API providers:

  • Amadeus for hotel inventory and pricing
  • Yelp Fusion for restaurant discovery
  • OpenWeatherMap for hyperlocal forecasts
  • We implemented a smart caching system using Redis that reduces API calls by 40% while maintaining data freshness. The cache is strategically invalidated based on each data type's volatility - weather updates every 30 minutes, while hotel prices refresh hourly. For premium users, we added a real-time mode that bypasses caching entirely.

The Fetch.ai integration represents our most forward-looking component. Three specialized agents operate in concert:

  • The Price Tracker monitors rate fluctuations
  • The Alert Bot manages user notification preferences
  • The Validator cryptographically signs all deal alerts
  • This decentralized approach ensures users receive tamper-proof notifications directly to their Fetch wallets. Our benchmarks show the agent network processes alerts in under 800ms, with each transaction consuming minimal gas fees (0.0015 FET on average).

Challenges we ran into

We have an api problem, a sort of war, an api war with Zomato, Amadeus. Some of the APIs used aren't the clearest and easiest to understand. Repeated Testing of the front end as one function break led to other things breaking, which made it very annoying.
(Also, the API keys in our code are still present as we didn't have enough time to remove it before pushing since we were cutting too close for submission - so we will shutdown them down by end of this competition instead as we do not want to update the code after submission and risk ourself getting removed from the competition.)

Accomplishments that we're proud of

It took us a while to complete the project, it was even near deadline, and as we live in Thailand, we had to stay up a couple of nights, as we had to rush our projects since we are still university students, and our exam dates was near deadline so we had to focus more on the exam than the hackathon but once we were done with our exams. We were dedicated completely on the hackathon and finished it just in time before submissions.

What we learned

Through the development of GlobalMate, we gained invaluable technical and philosophical insights that fundamentally changed our approach to travel technology:

  1. The Illusion of API Stability: We learned the hard way that even established APIs like Zomato can suddenly become unavailable, forcing us to completely rebuild our restaurant module within 72 hours. This taught us to:
  2. Implement abstract adapter patterns for all third-party integrations
  3. Maintain parallel API providers for critical services

  4. Develop robust fallback mechanisms that degrade gracefully

  5. The Psychology of Travel Planning: User testing revealed unexpected behavioral patterns: Also, the restaurant function works, but the API has to be obtained as a YELP partner.

What's next for GlobalMate - AI Assistance for Travel Guide

We can use feedback from the users, and based on that, the most requested feature will be added to the AI for future iterations.

Built With

Share this project:

Updates