Inspiration
I, Izabel Jipson, volunteered as a trained peer support specialist at Sean's House, a mental health center located on the University of Delaware's campus. Peer support specialists are different from therapists in that we are trained to use our mental health struggles to validate the emotions of visitors and give them local tools and resources to aid in their recovery. We are not intended to replace therapists but rather offer a baby step to helping people feel more comfortable seeking professional help. I noticed that while these services were useful, our hours were limited to the daytime from 8 AM to 8 PM when oftentimes more students begin to spiral late at night without access to the same services. This inspired me to work with my team to create PeerEI, a mental health tool intended to assist not replace mental health professionals.
What it does
This chatbot is trained on the mental health resources specific to Georgia Tech as well as how to help the user develop a Wellness Recovery Action Plan, aka a WRAP Plan, which is a standard peer services tool used to help guests build call-to-action plans to aid in their recovery and give them a better sense of control. Additionally, five unique characters represent a wide array of students with different types of struggles at Georgia Tech. For example, our athlete character is best suited for athletic-related mental health struggles or time management issues, whereas the Greek life character is best suited for social and relationship struggles.
How we built it
We trained the OpenAI API to design our chatbot in Python and used prompt engineering to fine tune specifics regarding each character as well as what a peer support specialist is meant to do as a whole. The frontend messaging system was built using Flutter for the interface and the resources page uses basic HTML and CSS. Additionally, we connected everything together using a Flask server.
Challenges we ran into
Some challenges we ran into were integrating the API into our project and connecting all the moving parts through the Flask server. Additionally, fine-tuning the chatbot was a little challenging in getting the AI to sound like a college student and to keep the responses brief as though it were a text chain.
Accomplishments that we're proud of
We are proud that we have a final product that is fleshed out and interactive. Additionally, the fact that we have created a project for social good has the potential to have a real-world impact.
What we learned
We learned a lot about how to integrate APIs and create a narrow but targeted goal for our chatbot. We also learned about how to use Flask servers and fine-tune the Open AI LLM.
What's next for PeerEI
Although PeerEI is currently trained specifically to the resources and students of Georgia Tech this is a flexible model that can be adapted to other universities and is fully scalable.
Log in or sign up for Devpost to join the conversation.