Destressify

Uses machine learning & OpenAI to suggest activities to destress. The user will answer questions or put a prompt into GPT 3.5 to quantify their mood which will then recommend an activity.

Comment

Inspiration

Students at WPI deal with lots of stress. Something that a lot of our chapter found in common was that because WPI is so fast-paced with our quarter system, we are often constantly moving up until the deadlines to get assignments done. After an assignment is done, however, we can usually find time to relax. However, something that happens to many of us is that during this time is that we get a "weird" feeling not doing work - like we should be doing something active or productive. The main idea for this app is that it will recommend an appropriate destress activity for someone to do, when they can't think of what to do. This could happen for a number of reasons - they don't feel like doing any of the things they usually do or that they're too stressed out to even think about what would be relaxing.

What it does

Given user input recommend an activity to do to destress

How we built it

We used Python, Tensorflow and Sklearn (for train-test split) to develop our machine learning model, which is a simple softmax regression. To generate the mock dataset we used mockaroo.com. To integrate GPT-3.5, we used the OpenAI API.

Our machine learning model takes a series of quantified answers from the user to produce an activity recommendation. An alternative input to answering these questions "manually" (ie. answering "how energetic am I? on a scale from -3 to 3), the user can put in a prompt into GPT-3.5, which will that response into a quantitive value.

Challenges we ran into

How to implement OpenAI API, finding good data

Accomplishments that we're proud of

Developing a strong understanding of machine learning and learning how to use tensorflow. Also making calls to OpenAI's API.

What we learned

We learned how to utilize OpenAi API into our project, although incomplete. Learned how to use Figma. Learned how AI can be more emotionally aware compared to humans through some research shown here: link

What's next for Destressify

Reinforcement learning for rating activities Connecting front-end and back-end Creating an accurate, non-random dataset Adjust user questions based on established psychology theory Integrate existing online services for activity recommendations ie. built-in Wordle, display nearby restaurants

Built With

  • figma
  • openaigpt-3.5
  • python
  • sklearn
  • tensorflow
Share this project:

Updates