Mentony

Problem & Motivation

This project proposes to take publicly available LlaMA 2 models and depth test them to see if the model can have an empathic supportive text conversation with a human. Conversations should show three things.  A basic level of empathy, expression that a described situation might be tough, or that a day was good. A response to hot words that signal crisis and need to be elevated to a hotline or nationally available service. Lastly, some indication of awareness of different larger cultural groups and their experiences in the United states.

To test cultural competency, Black American experiences and perspectives will be shared with the bot to see how it engages with those conversations. This particular cultural group has been chosen for several reasons. One in every three black people are receiving care for their mental health (Watkins, Marcie, et al). Health inequities, historical trauma, access, and lack of cultural competency make it tough for Black Americans to seek care for their mental well being. 

The project team is composed of Black Americans, the cultural group is underserved, and has some broad unified experiences that they may want to talk to someone about. For example, many African Americans have heightened anxiety about even normal interactions with police and may want to talk to someone about it but might not want to end up in a feedback loop with a friend. This project aims to educate and lighten the emotional load by making a web app for the black community to leverage.

Our Solution

A web app that promotes mental health awareness while being culturally sensitive. In addition, we want to test its sensitivity using experiences and inputs from the black community, with the end goal of encouraging users to seek professional support. Thus, bridging the gap for mental health treatment. The web app will host a chatbot that will offer emotional support for those that need it as a buffer. The goal being to help those who are in a momentary dark place, do not have access to mental health care, and/or those who do not seek care due to stigma, inequality, or other reasons. We tested efficacy of the MVP with a usability test where feedback was gathered from individuals from the black community. We used multiple metrics from the generative AI community including the chatbot test and acute eval.

Key Learnings & Impact

• Managing resources for state of the art technology
• Leveraging LLMs (Large Language Models) to create technology to assist with emotional wellness
• Understanding how LLMs deal with cultural sensitivity and cultural competency

Acknowledgements

We want to thank our capstone instructors Fred Nugen and Korin Reid. Our instructors helped us focus the scope of our project and helped us organize our goals with thoughtful feedback. Also, we would like to thank Shardel Hodge Matthews MA, LMFT who was our subject matter expert. She helped us prioritize features that were important for our potential users. In addition, she educated us on the role our chatbot should play in users' mental wellness journey. Her feedback shaped the outcome of the entire project.