Guardian Angel

Main home monitor page with baby monitor screen
Lullaby page where parents can record and save audio files of them reciting lullabies
Setting page for toggling light or dark mode

Inspiration

After meeting and getting to know everyone in our group, we discovered that one of our group mates is expecting a newborn child in 3 months. Seeing how everyone wanted to develop something that truly help people within their day to day lives, we decided to help those who were about to welcome new life into their lives properly manage and balance their own lives. So that is how Guardian Angel came to be.

What it does

Guardian Angel is an all in one suite for all your baby monitoring need. Guardian Angel is a software service that allows for instant WebRTC connection between your personal device such as your phone or desktop and your device monitoring the baby. Guardian Angel allows parents the chance to go do chores around the house while their children are sleeping or laying around and if crying is detected, Guardian Angel alerts them via text notification that their baby is crying to which they have the ability to unmute their device and immediately in real time talk to their baby. Guardian Angel also has many other features such as a secure login system, the ability for parents to record and store lullabies, and the ability to generate and read out personalized and tailored stories. With the feed page allowing us to turn any old device to enter a state ready for peer to peer connection for baby monitoring.

How we built it

Guardian Angel relies on a WebRTC connection, to establish peer-to-peer connection between your device being used as a camera (such as an older phone) and the parent's device (phone, laptop, etc.), while connected it leverages a trained from scratch deep learning neural network utilizing an atom optimizer in order to classify sounds as regular day-to-day actions/sounds or whether it is a crying sound. Leveraging a dataset comprised of about 900 audio files with varying lengths and complexities labelled under crying sound or not crying sound. Using the YAMnet deep net to create embeddings on the audio files for training. The secure login system was made via the Auth0 service to securely authenticate and store user data, while our generate story feature leveraged Google gemini API and then eleven labs to read aloud the stories, and using MongoDB to store personalized lullabies in the cloud. With twillo being used to send an SMS message to the parent when crying is detected.

Challenges we ran into

A hard part of this development was finding an appropriate/balanced dataset for training our crying prediction model. Furthermore establishing the peer-to-peer connection between the device being used as a camera and the parent's device proved hard to do. With authentication and properly being able to store lullabies proving hard too. However with tenacity dedication and alot of documentation we were able to push through these challenges.

Accomplishments that we're proud of

We were proud when our trained from scratch AI model was able to properly predict and distinguish between crying and regular background noise. Another accomplishment we were very proud of was being able to establish the peer to peer real time connection between our device monitoring the baby and then our parent's device.

What we learned

We learnt the importance of data processing for ai training and choosing the right dataset, furthermore we learn how to establish a peer to peer connection using webRTC and establish our own peer to peer provider, also were able to deploy the project to be used on the internet.

What's next for Guardian Angel

Future aspirations for the guardian angel team is to go beyond software and then dive into hardware too, building a standalone baby monitor device that leverages our software to detect further information such as the temperature of the room, humidity of the room and noise level of the room. We also wanted to develop a computer vision model to detect if the baby was sleeping or moving in a way that would be dangerous to their health. Finally we would live to have a feature that records down times and