esai | Devpost

Inspiration

We were inspired by our interdisciplinary backgrounds in psychology, data science, computer science, and engineering. It's not easy to put labels on mental health issues, so we wanted to find a way to search for strong mental health resources without knowing the term first. Specifically, we were unimpressed with current searchable resources (webMD for example), so we wanted to find a reliable source. No source is more reliable than the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, known as the DSM-5, the standard for psychiatrists and psychologists in helping aid mental health diagnoses.

What it does

esai parses through the entire DSM-5 with semantic search. Basically, a user can roughly describe how they're feeling - such as "I'm just not having a great day", and esai will source both the name of any applicable mental disorders and the DSM-5 reference material, along with diagnostic attributes.

For physicians, esai will show applicable analytical information about patients - giving it a wide variety of use cases from being a personal mental health aid to an enterprise tool for professionals.

How we built it

esai uses OpenAI's embedding engines, GPT-3.5 Turbo, Pinecone, and Langchain, built on Next.js.

Challenges we ran into

A significant issue was configuring Pinecone to function with our vector embeddings. We also struggled to find ways to excerpt the information from DSM-5 pdfs.

Accomplishments that we're proud of

We are proud that we learned how to use Pinecone, LangChain, and the OpenAI API, along with building a fairly complex algorithm for scraping the DSM-5 from an HTML version of the DSM-5, accessed through the UC Davis library VPN.