Unison

• 

  Unison's Logo: Stronger Together

Inspiration

Mental health is a huge problem, especially with high school students, and we took notice of it around our community. We also observed that parents often showed concern for their children but discussing mental health with adults was considered taboo. We wanted to create a space where teenagers could anonymously share their feelings while still providing a way for parents to have insight into how their children are doing.

What it does

Unison uses advanced NLP to analyze conversations between kids and an AI therapist and then communicates the mental health of the child to the parents in an easy-to-view parent dashboard. Furthermore, Unison uses OpenAI's GPT-3 models to provide suggestions to both the children and parents as to how to improve their mental health.

How we built it

Unison is coded in React-Native with a backend Firebase Node.js server managing user authentication, cloud functions for daily notifications, and a database to keep track of families and their children. The NLP model was trained in Python using Tensorflow and is a fine-tuned BERT model.

Challenges we ran into

Machine learning is an extremely complex topic that required numerous different attempts of creating a reliable model. Additionally, connecting and securing data between Python and Javascript backends to a React-Native application provided more difficult than originally anticipated.

Accomplishments that we're proud of

We are extremely proud of the user interface and robust architecture which we were able to produce in just over 36 hours' time. Not only do we have a multi-language application but we also incorporated APIs, Neural Networks, User Authentication, Scheduled Push Notifications, and more.

What we learned

Our frontend developer gained a lot of experience and knowledge in React-Native which is extremely useful as it can be compiled to both iOS and Android. Meanwhile, our two backend developers learned a lot about the ins and outs of text classification, sentiment classification, and NLP models.

What's next for Unison

With more time, the UnisonAI Team can improve the neural network by finding more accurate and representative data as well as collaborating with mental health professionals to provide better insight into how we can detect and act on improving mental health before it becomes an irreversible issue.

Built With

• firebase
• natural-language-processing
• node.js
• python
• react
• react-native