BestFriend

BestFriend is a mental health app that aims to alleviate stress during times of high traffic on the suicide hotline with a conversational bot whilst also providing a more casual therapy option.

Comment

Inspiration

According to the World Health Organization, over 700,000 people commit suicide each year. Crises hotlines are as necessary as they've ever been in combating suicide incidents. However, incoming calls to the U.S. National Suicide Prevention Lifeline have nearly doubled since 2016, and in 2021 about 330,000 calls were abandoned before a caller could receive help (New York Times 2022). We want to address this through BestFriend's key feature and retain as many callers as possible.

What it does

BestFriend's main goal and feature is to serve as a text-based immediate responder to Lifeline callers when they're unable to be immediately transferred to a counselor. Additionally, it has a therapy chat feature that speaks to a user like a therapist would. Lastly, it provides users access to therapists in their local areas and provides a rating option.

How we built it

We fine-tuned a GPT-3 text-generation model to act as a conversational bot. We used the vader-lexicon model to train our Natural Language Processing module. Our back-end was created with Flask and hosted on Google Cloud Platform using a Docker container. It would make calls to the OpenAI API and fetches responses from user requests. Our front-end was a Flutter framework, which allowed us to deploy the app to multiple environment systems, allowing for a wide user-base.

Challenges we ran into

We initially attempted to train our own AI model using a dataset from Council Chat, an online chat-therapist service, whose data was readily available. However, it was both unethical to source that data and overall difficult to work with effectively. Setting up a back-end deployment pipeline with Google Cloud Platform/Run proved difficult as well. Since mental health is a very complex problem-area, it was difficult to fine-tune our AI to give sensible, sanitized, and ethical responses.

Accomplishments that we're proud of

We all came in with relatively little experience in deploying apps in Google Cloud Platform. Additionally, half of our team had never previously designed a front-end user-interface. Overcoming these difficulties was a valuable learning experience for all of us. We consider our AI system to be the most advanced that we've had the opportunity to build, and seriously view the potential of our system as a product and the impact it can have on people's lives.

What we learned

We've gained a renewed respect for human call-center operators and the heavy responsibilities they take on. On the technical side, we got experience using and fine-tuning AI, as well as developing cross-platform user-interfaces.

What's next for BestFriend

Although we have no plans to replace human operators in the counseling space, we feel that expanding BestFriend's services could further alleviate the current system of overloaded call-centers.

**

Share this project:

Updates