MySafetyNet

Inspiration

Mental Health has been a very important topic during the last years, annd even though we are getting more and more people to get help much easier, there remains a small portion of those who dont feel confident enough to talk with a person about their struggles.

What it does

MySafetyNet is a private space for our users to talk about their feelings, where an AI will be tracking their mood with some questions designed for the user to feel free, but also keeping a record of questions that a psychiatrist would ask to diagnose a mental disorder like depression.

How we built it

We applied transfer learning to MobileBERT, a NLP state-of-the-art model, with the purpose of assesing Sentiment Analysis (natural language task) on user´s text input through a mobile device. Before we were able to get inferences through a mobile device with the trained model, we needed to apply conversion and quantization using TensorFlow Lite. We integrated our model into an iOS app using Xcode, with the help of cocoapods and homebrew.

Challenges we ran into

One of our first challenges was trying to find a good dataset that fiitted our needs, we trained twice the model before seeing good results. The whole app building part was challenging since we didnt had a lot of experience with it, especially the part of inserting the trained model.

Accomplishments that we're proud of

The whole proyect feels like an accomplishment, comming up with the idea, training such a NLP model based on sentiment analysis, applying conversion and quantization to it, and then importing it into an app felt unrealistic not so long ago.

What we learned

We sure learn a lot new concepts and techniques, but the most valuable thing we've learned this weekend is how far we can go with our capabilities, it surely is a confidence boost!

What's next for MySafetyNet

Since this is a very short-time proyect and its touching critical aspects of people's life and health, we would like to review the questions asked, as well as the analysis done from the answers, which require both medical supervission and better learning methods that can decide how to assess the person in a more sensible way.