LoveScout

Kyuubey's evil sales pitch
WIP Screenshot of the app
Performance and upcoming features of the CNN

Valentine's is just around the corner and you're a little short on date ideas? LoveScout is here to guide you through an enamoring promenade through the best local attractions with the partner of your choice.

What it does

LoveScout is a first of it's kind date enhancing app, which prompts the user with personalized missions, or scouts, to complete during their romantic encounter. Taking the form of a scavenger hunt, the participants will be prompted to discover nearby attractions and activities and immortalize the moment with a picture. From temporary art exhibitions to delicious food fairs, LoveScout will always be on top of the most recent happenings in your city.

How we built it

LoveScout is a python based Kivy application that uses machine learning to complement it's features. Our use of machine learning is twofold. First, we mainly use a Large Language Model (LLM) to create the date prompts, which allows them to be randomized, customizable and infinite. In addition, a Convolution Neural Network (CNN) is the main technology behind the image recognition after each mission, to confirm if the prompt was completed or not. The main advantage of using Kivy to create our app is that it is easily converted to an APK package for distribution across android platforms. This makes it practically ready for a full release at any time.

Inspiration

At the crossroads of GeoGuessr and dating apps, LoveScout was created as a fun way to get to know your partner by exploring the best your environment has to offer. Struggling to find date ideas for the upcoming Valentine's will be a thing of the past!

Challenges we ran into

We had very little experience with machine learning before starting this project, but installing every dependency such that they are compatible was the most time consuming process of development. It was also our first time working with Kivy, which accounts for many headaches on it's own as well.

Accomplishments that we're proud of

The implementation of machine learning into an app was a daunting step we are glad we took. Indeed, taking advantage of such a powerful resource has allowed us to create a highly marketable app that we are proud of.

What we learned

We got really comfortable with python in a creative setting, which is unusual for mere science students. Our implementation of machine learning was also very insightful many hours were dedicated to researching the best models for our goals.

What's next for LoveScout

Our main problems were with the accuracy of the deep learning convolution network, which could be more accurate. This can be accomplished by training it on a better and more precise dataset. As it turns out, it's hard to categorize everything in Montreal. Additionally, instead of a sequential model, a more complex functional approach to the CNN could yield better results due to the higher complexity and variety of the data. Of course, we would also like to make our app more visually attractive, after all, shouldn't an app about dating be attractive?