Vetricle

An app designed for veterans to arrange meetups and chat with friends. Vetricle because it keeps them close at heart.

Comment

Inspiration

When returning from deployment, veterans often feel estranged from society. Gone is the rigorous structure, clear-cut purpose, and sense of camaraderie found within in the armed forces. Our team aimed to tackle this problem in order to give back to those who selflessly serve our country.

What it does

The app matches veterans based on their interests, allowing them to chat and participate in activities. For example, veterans who are interested in getting their hands dirty with hardware or are curious about software development would be able to find a community of likeminded individuals. Furthermore, making connections or creating something tangible with others can bring purpose to veterans' life.

A user can sign up with their email or authenticate with their Facebook account. Afterwards, the user may find events nearby which are tailored to their interests. They would also be able to see if any fellow veterans are RSVP'd. Vets may also have the option to host their own event. Based on similar interests and activity, veterans may get friend recommendations. Lastly, veterans who opt-in can chat or meetup with nearby veterans.

How we built it

The website was built using React.js and the app was created using React Native. The backend was created using Python and Flask. The backend authenticator was built using React.js with Google Cloud's FireBase and FireCloud database. Data in the form of JSON's from the app or website are run through analytic scripts on Python that call IBM Watson's Apache Spark models on the Watson studio to do data analysis to create friend recommendations for veterans based on similar interests. If not available, it calls upon collaborative filtering methods to find user to user similarities. If we had more time, we would have incorporated a three-layer Neural Autoencoder made up of LSTM Recurrent Neural Network units that would be made on PyTorch and trained on Google Cloud with access to GPU acceleration for more accurate recommendation analysis as the number of users grows.

Challenges we ran into

-CSS issues and formatting JSON's in a correctly

-Getting the chatting service to work at force

-Large dimensionality of the dataset (if the number of users is large, it can be challenging to do analysis)

-Lack of information on using recommendation systems for data that isn't dependent on a rating system

-Not enough time to run PyTorch autoencoder model.

Accomplishments that we're proud of

Designing a clean, cross-platform UI for the app, coming up with a consistent dataset for usernames and learning more about recommendation systems are all accomplishments we are proud of.

What I learned

We learned how to create an app using React Native which allows for simultaneous release on both iOS and android devices. The new members on our team also learned how to use Javascript, Adobe XD and utilize Git. We also learned how to do collaborative filtering and content based filtering which is heavily used in recommendation systems today as well as learning about deep learning techniques for recommendation systems today such as Restricted Boltzmann machines and Neural autoencoders.

What's next for Vetricle

We hope to increase the app's functionality and grow its userbase. We would also like to create a fully trained Neural Autoencoder model using a large set of data for more streamlined recommendations. Additionally, we would like to incorporate more factors for recommendations (i.e mutual friends).

Share this project:

Updates