Mental health screening Chatbot API

Inspiration

After winter break, the therapists at CU Boulder were flooded with more screening appointments than they could often handle. Because of this, these medical professionals reported high stress, and we were wondering if we could use AI to help mitigate such a crisis.

Additionally, we were shocked by an article about how a teen’s suicide after talking to a CharacterAI chatbot, and we were hoping to mitigate the adverse effects that AI has on mental health and move towards a framework of prevention.

What it does

Using text or audio input, the model uses natural language processing and leverages pre-existing AI chatbots to learn how to effectively respond to users based on mental health conditions that may be concluded from the model.

How we built it

Using an open-source NLP model, we planned to fine-tune the model using a public dataset to better analyze the likelihood of mental health conditions through analyzing text. In this case of audio input, we would have converted audio to a text transcript using an open-source speech-to-text model. We would have integrated the results into an API to be widely used by healthcare professionals and AI chatbots.

Challenges we ran into

Datasets were not as complete or ideal as we wanted (lack of mental health conditions for validation, most were from social media, etc.). We had difficulties utilizing the RoBERTa Model and figuring out how to fine-tune it using our data. It was hard to narrow down our project idea and it entailed a lot of discussion.

Accomplishments that we're proud of

Our first hackathon! :) Adapting and changing our project ideas when others weren’t working out. A pretty slideshow

What we learned

We learned about how to utilize the 24 hours that we are given in a hackathon and how to keep on persisting when a project isn’t working. We learned about sentiment analysis and how linguistic features are analyzed and extracted.