Inspiration : Safety and wellness of kids motivated us to come up with this idea.
What it does : When a sponsor wants to talk to a child who lives in a different part of the world, he initiates a conversation with Alexa and the voice gets converted to text, fed through a ML model which filters the inappropriate words, translates the filtered text to the kid's native language and speaks out through Alexa.
How I built it : Built with various AWS services such as Lambda, Sage maker, S3, Quicksight, Glue, Athena & cloud9. Built and trained a ML model using Python.
Challenges I ran into : Calling two levels of nested Lambda's and getting the output returned from the second Lambda to Alexa. Difficulties in incorporating external python libraries. Training the ML model in a short time. Configuring Alexa to a different language.
Accomplishments that I'm proud of: Developing the whole end-to-end prototype model using a plethora of AWS tools which are so distinct from each other.
What I learned: How to leverage Alexa in conjunction with translate API's along with machine learning models.
What's next for SafeCom : Extending the API to leverage Amazon Poly & Amazon Transcribe.