Aura GroupSolver

Create a party and invite your friends to choose your next destination without any conflicts
The chatbot will ask you for some details about your plans for your next trip.
Wait for every friend to agree and get travel suggestions - view them on a 3D map

Inspiration

Planning a trip with friends or family is always tricky—everyone has different expectations, different preferences, and different budgets. Why not find a fun and interactive way to solve this problem?

What it does

Aura GroupSolver is a tool that helps friends decide on their next travel destination in a fun way ! It uses an AI agent to make sure you’re all on the same page and alerts you if there are any points of disagreement, such as price, dates, destinations, or the type of trip. Finally, it presents possible destinations in an engaging way.

How we built it

We run a local Gemma 2B AI model that powers an interactive agent. This agent engages with users while retrieving data from a Firestore server to monitor other users’ decisions and determine when a consensus has been reached. To provide optimal travel recommendations, we integrate the Skyscanner API to access real-time flight prices. Finally, recommended destinations are displayed on a dedicated page built using Three JS, R3F, GLSL, React, and Drei.

Challenges we ran into

The biggest challenge we faced was deploying the AI model locally. Most of the models we tested produced good results but were too resource-intensive to run on our modest laptops. The less resource-intensive models, on the other hand, yielded less satisfactory results and produced numerous errors in the AI agent’s responses. We therefore had to find the right balance.

Accomplishments that we're proud of

We're really proud of the web app's user interface, but also of having deployed the AI model locally for a more environmentally friendly solution and enhanced privacy !

What we learned

Every team member was able to learn new things at all levels, whether it was how to use GitHub, how to use generative AI to develop a web app, how to deploy Docker containers, how to deploy an AI model locally with Ollama, how a React frontend interacts with a Python backend, and so on.

What's next for Aura GroupSolver

One interesting approach would be to use more attributes to determine whether trips conflict with one another. It might also be worth exploring the possibility of deploying this system using a P2P protocol over the internet (rather than just locally, as in our prototype).