The problem Comfy solves

We created Comfy, your personal healthcare companion. Comfy is a Mental Health Tracker-cum-Song Recommender System integrated inside an android app and a website. Inside Comfy, we created a chatbot, Alexia. The user can converse with Alexia, just like he’d talk with any other human. Alexia analyzes the responses to generate counter-responses.
Alexia also runs our machine learning model over the user’s responses to detect the user’s mood via Tone Analysis. As per the tone, the model returns a JSON data containing the emotion. Using this data, we query on the Last.fm API to recommend user songs as per his mood and taste. We also show the user a list of activities that would help him to make his day better.
Comfy also authenticates users via the Auth0 authentication.

About the ML Model :
We obtained the dataset from Kaggle. The deep learning model we used for training was CNN combined with word2vec. It gave an f1 score of 74.56 on the unseen test dataset. For deployment, we’ve used Flask.

Challenges we ran into

We tried hosting the model on Heroku but the size was too large. We tried fetching data by running it locally, but android doesn't request data without an SSL certificate. In the end, we did make our whole website and ML model with Flask which worked but until then UI wasn't good enough.