Share-On

• 

  title slide

How we built it

To achieve this chatbot, the Retrieval Augmented Generation architecture has been utilized. We have combined Open AI’s latest Gpt 3.5-turbo LLM with a database of relevant mental health info. Using the Llama Index library, our code uses different tools to process user input and find relevant data to synthesize a more nuanced and mental health specific response to user’s queries. To elaborate, the user's input is used to search a vector database of embedding that is associated with different data chunks in the database. Once a relevancy search is done, both the data and prompt are fed to the LLM to generate a response. The inputs are also collected, while still not collecting personal info, to observe trends and patterns that can be advocated for. This application has been deployed through streamlit onto a website that is accessible on the web, making it easy to use for students.

Challenges we ran into

The biggest challenge in building this application was taking into consideration the cost and effectiveness of the components we were going to use. With a tight budget, we tried out different options for models, deployment, and the website platform to find one that would work best. In the end our application was running successfully while still not draining our resources. A successful component of our chatbot was how we were able to integrate it into our website, so that users can access the chatbot and ,at the same site, info about our organization with ease.

Accomplishments that we're proud of

A successful component of our chatbot was how we were able to integrate it into our website, so that users can access the chatbot and, at the same site, info about our organization with ease. We were also incredibly proud of the fact that as a team, we all have similar vision with our chatbot, we all hope to put a step forward in changing the face of teen mental health. As a team, we have created chatbot for reasons far beyond glory, but rather because of our unfortunate exposure to the obviously flawed teen mental health system. Instead of giving into the stigma, we have combined forces to turn our unfortunate experiences into something great, so that the next generation would not have to live through such strong stigma and inaccessibility to mental health care.

What we learned

What we learned from this process was how much of it was trial and error. We found ourselves deviating from our original plan often and having to improvise. But this improved our critical thinking and problem solving skills, along with making our chatbot more advanced and thorough.

Built With

• amazon-web-services
• chatgpt3
• llamaindex
• openai
• python
• rag
• streamlit
• vercel