Inspiration
My inspiration for creating this mental health group chat for youth came from my personal experiences with therapy. Having grappled with the challenges of speech impediments and understanding the emotional toll it can take, I recognized the importance of providing a safe and supportive space for young people to connect, share their experiences, and access resources to navigate their mental health journeys. I have gone to professional therapists since elementary school, yet I have seldom felt entirely comfortable with them. They were kind people doing their best to help me, but as they did not share my condition, it was hard to open up to them. When I got older, I learned about this local support group with other stuttering youth and decided to attend. It was an incredible experience. I could genuinely open up and connect to people who had gone through the same things I did. One integral part of this support group that made it so inspiring was the presence of a college student with a stutter. He was slightly older than the rest of us, had come to accept his stuttering, and, in some ways, overcome it. He was much more open about his difficulties and experiences because of this, and he was able to guide and lead the conversation like a therapist would while still connecting with us due to being close to us in age and sharing our condition. This presence of an individual who has overcome their challenges, is willing to lead the conversation, and won’t be seen as judgemental is critical to these events. However, these groups are sometimes hard to find, and it can be hard to find these individuals. That is why I created MentalHealthYouthSolutions. This is an online chat platform with an AI moderator who will take the place of that leading individual. It will be able to lead the conversation while not being seen as judgemental due to its silicon nature.
What it does
YouthConnect allows users to create accounts and directly chat with others or form group chats with many different users. They can form groups centered around a common topic, such as speech impediments, depression, substance abuse, and more. The users can add Thera, the Large Language Model Therapist/Moderator, to this group. Thera would lead the conversation, help break the ice, and overall elevate the conversation by asking questions and providing positive feedback. As a transformer-based model, Thera can effectively use conversation history to do an optimal job. Users can also chat with Thera alone.
How we built it
MentalHealthYouthSolutions uses React JS and Chatengine to create user accounts and allow them to chat with each other, Thera the Chatbot, and form group chats that can include Thera if they wish. Thera is a Llama2-based Large Language Model run in Python using the Pytorch library.
Challenges we ran into
When creating this project, I had to optimize, clean, and process the text passed and returned from the model to ensure that it would properly understand and respond to the question. I also had to interface a javascript backend with the javascript frontend cleanly, and it took multiple tries to get a consistent interface. It was incredibly difficult to run the LLM locally in a way that was accessible by code, but we eventually found a way through a lot of trial and error.
Accomplishments that we're proud of
I am proud of creating a complete frontend and backend chat room alongside a working Large Language Model interface. I had to improve my knowledge of front-end programming javascript and had to continuously work on processing the textual data for the Large Language Model. I am also really proud of managing to interface the LLM directly into the website. For past projects, I have struggled with AI because it is so resource-intensive. During this project, I found a way to run the model on my CPU so that it could directly interface with the website.
What we learned
I learned how to create an aesthetically pleasing and completely functional chatroom using JavaScript and effectively run a Large Language Model interface using Pytorch.
What's next for MentalHealthYouthSolutions
The large language model requires lots of processing power, yet due to limited resources, I can only run it on Google Colab, so in the future, I would like to get access to powerful cloud computing servers so that the AI model can be integrated within the chatroom. I would also like to expand this to the IOS and Android app stores, as it is currently available only as a website. Additionally, I am currently running the LLM on my laptop, which means that is really slow. Servers are expensive, especially those that can run such a model quickly, but if I were to get funding in the future for a server, the model would run much faster.
Log in or sign up for Devpost to join the conversation.