PrivateTherapist

Inspiration

In-person therapy can be inaccessible; online therapy can be more accessible but not private (as you may not know where your data goes)

What it does

-downloads model to the browser (tested on Chrome) -user inputs message

• system guides the user using as in a therapy session
• if it determines that it needs external resources (using Senso), it ask for confirmation; if granted it interacts with another agent (using Apify) to scrape relevant web content.

How we built it

• We used a lightweight llama model to run locally -We used Senso to determine if more info is needed -We used Apify to connect to another agent to get that info (using website-content-crawler) ## Challenges we ran into -The model is still to big for some devices -The inference is too slow without GPU support
• Apify's web content crawler might be too slow for real-time responses ## Accomplishments that we're proud of -Senso was surprisingly good at classifying the case severity without much context ## What we learned It's possible to enhance in-browser private features with external agents without fully compromising privacy ## What's next for PrivateTherapist
• broaden device support
• search for useful agents in Apify
• improve the classification using Senso

Built With

• apify
• javascript
• python
• senso
• vizcom