Inspiration
Mental health issues and disorders are a huge problem in our society. According to the CDC, over 50% of American individuals will be diagnosed with a mental health illness or disorder at some point in their life. If these mental health conditions worsen, not only do they put the individual's life at stake, they might even cause a decrease in worker efficiency! (joke) Additionally, many want to take steps to improve their mental health but do not know how to track or treat it.
What it does
OskiHealth is an android application which aims to improve the mental well-being of individuals. It does this by automatically tracking the user's mental well-being, as well as providing an avenue for users to get help, via a few main avenues:
- Daily check-in: Every day when the user opens the application, they have to input their feelings for the day (0-4 stars). We use this to analyze their well-being over time.
- AI Well-being Chat-bot: Sometimes, there are things you cannot tell anyone else. In this case, you can talk to our custom GPT 3.5 generative AI model, tuned for assisting one's mental health journey, that can act as a confidant and advisor.
- Sentiment analysis of messages: The app includes a messaging platform that you can use to message your friends. This will keep users motivated to use the app. However, the more important feature is that the messages sent are processed via an NLP (Natural Language Processing) model to determine the positivity/negativity of the messages, which is another component of our well-being analysis. No commercial application has this feature.
- Well-being analysis and analytics: With all the data we've gathered, we can show trends in your mental health and how your mental health journey has been since you've started using the app, via elegant charts and graphs. This makes tracking well-being hassle-free.
- Get Help: If we detect that the user's mental health might need help, we give them a list of actions they may do to improve their well-being, as well as articles they can read to further help themselves. Additionally, after a cash donation, they can access professionals such as psychologists and therapists, who will analyze their data and talk to them directly. This fee helps to compensate the professionals as well as pay for the upkeep of the service, which makes this app practical and scalable.
How we built it
We used Android Studio to build the Android application, following Google's own Material 3 design guidelines for accessibility and aesthetics. It connects to our web server, written in blazingly-fast memory-safe Rust using the Rocket library, which does the majority of the networking and message transfer, as well as interfacing with the OpenAI API which runs our fine-tuned GPT 3.5 model which we optimized for mental health. However, sentiment analysis is run on a Flask server and PyTorch using a pre-trained inference model from HuggingFace, which was trained on Twitter tweets which means modern language will be understood by this model.
Challenges we ran into
The free AWS instance we used to host the Rocket and Flask servers did not have enough RAM to install the dependencies for sentiment analysis. Fortunately, we could create 2GB of swap space which helped us get through the installation process. This is also our first time using the OpenAI API, so we were unsure of how to tune the model to our liking, and we had to liaise for 5 dollars to pay for the service. It is also our first time using graphs and charts in Android, so it was difficult to configure the library used and to make it work.
Accomplishments that we're proud of
- Functioning and surprisingly accurate sentiment analysis of messages
- OpenAI GPT 3.5 integration
- Cool and funky graphs
- Working analysis of data
- Nice looking UI
What we learned
- How to use sentiment analysis and custom GPT 3.5 models
- How to plot graphs in Android
What's next for OskiHealth
We could start by getting help from a real psychologist to fine-tune our data collection methods and analysis algorithms. Afterwards, we can pay that psychologist to be our in-house psychologist and help those who pay for the premium service while we launch the app. If the demand is high, we can scale up and get better servers and professionals.
Log in or sign up for Devpost to join the conversation.