Nova | Devpost

Inspiration

With extraordinarily high healthcare costs in the U.S., it is very important to cut as many costs as possible. A large portion of this waste comes from improper drug use and resulting hospital readmission. In particular, the elderly are not very good at keeping track of their medications and often forget certain doses. While solutions to this have been built, many require a computer, smartphone or tablet. The elderly often don't know how to use these devices.

What it does

The advances in voice control offer an interesting method to solve this problem. While a certain amount of expertise is required for a computer or tablet, devices such as Alexa require no such expertise. We built a skill for Alexa that helps the elderly to keep to the proper dosage schedule for their medications by providing an easily accessible platform where doctors can set the medication schedule and patients can be reminded of it. This provides both an increase in overall system efficiency and a decrease in the likelihood of hospital readmission due to improper drug use.

How we built it

We used the Flask-Ask library alongside Amazon's standard Alexa SDK to create the Alexa skill and the main server. The client was written in HTML5 with jQuery, Bootstrap, and moment.js.

Challenges we ran into

Amazon's documentation seemed largely focused on the protocol level documentation, with literally no actual code outside of examples. In addition, the Flask-Ask library is only referenced in Amazon's communications via the blog of an employee, rather than any actual reference or call-out in the documentation.

Furthermore, there apparently isn't a data type (or "slot type") for an English sentence. A musician, yes. A "municipal service," yes. A British first name (as distinct from an American first name), yes. An English sentence, no.

Accomplishments that we're proud of

Alexa produces sound in response to speech.

What we learned

Maybe we should build our own hardware and just use the machine learning people's machines to train a speech-rec neural net on instead of dealing with this...

But it looks like the Amazon SDK has a pretty huge English corpus, so...