Smoke Alarm

Inspiration

Burnout is a prevalent problem in Singapore, with an estimated 37% of Singaporeans suffering from it. As the "burnout generation", we have seen how burnout can kill motivation, damage mental health, and affect the ones around us. However, if caught early, burnout is much easier to deal with and prevent. Unfortunately, the early stages of burnout often involve a surprised state of heightened motivation - resulting in many not realizing that their stress levels and negative emotions are continually increasing until it is too late. With smoke alarm, we have created an early warning system so that we can avoid people experiencing complete burnout and take a break when necessary.

What it does

Smoke Alarm is a Twitter bot that monitors your mood through your tweets. When you follow the account, the bot begins to track your tweets and generates an emotional history of these tweets using a state of the art deep learning NLP model. When stress levels are consistently rising or high over a given period of time, Smoke Alarm's twitter account will send a message to the user to provide early warning for the potential of burnout, urging the user to take a break before burnout really occurs.

How we built it

Smoke Alarm was built on NodeJS and MongoDB hosted on Heroku, using Python's python-twitter API wrapper to interface with twitter. When new accounts follow Smoke Alarm, they are automatically added to a list and their most recent posts are analysed to generate a general report which is then sent to them for them to better understand their current mood. Further analysis is performed periodically in order to track their emotional history and determine if they are at risk of burnout. The emotional analysis is performed with a roBERTa NLP model trained on the TweetEval emotional classification benchmark and hosted on Huggingface Transformers.

Challenges we ran into

Some of the challenges involved trying to set a threshold for what constitutes an increase in negative feelings. It was also difficult to find well trained model that were specific to stress levels and burn out, and we resorted to a more general model as a result.

Accomplishments that we're proud of

As developers who were fairly new to backend development, we were quite happy that we were able to integrate machine learning into this project and get everything up and running within such a short span of time!

What we learned

We learnt just how difficult back end development is - making sure all the moving parts fit together and ensuring that our application was reliable and ran smoothly. We also learnt that there truly is an API for everything - it definitely would have been quite a bit of trouble to deploy our machine learning model ourselves.

What's next for Smoke Alarm

We hope to add a front-end application that allows users to submit their twitter handles and generate wellness reports and graphs for them to better understand the recent state of their mental health. We also hope to expand to more social media platforms like Instagram or Facebook for a more comprehensive overview of a user's mental health.