The inspiration for our project is a combination of technological passion and personal experience.

Tzahi and Kevin are part of an engineering team which specializes in exploring and implementing cutting edge technologies to solve business problems. Machine Learning based technologies provide a uniquely effective toolset for addressing the problems of those living with type-2 diabetes. The team has advanced experience in implementing voice based and chat bot solutions.

Kevin's uncle, Wayne, suffers from type-2 diabetes and struggles to manage his blood glucose levels. As a result, Wayne recently lost the lower half of his leg and is struggling to maintain the health of the other. Our findings indicate Wayne is not alone is his struggle to effectively and comfortably manage this condition.

The personal experience of one of our team members inspired us to contemplate the consequences of applying our skill set to the problems of those suffering from this chronic illness. Our solution is a voice & sound based guardian angel (Features will be described in the following section).

What it does

General Description of Guardian Angel Guardian Angel is a voice based application which lives on a device such as (Amazon Alexa, Google Home, Siri, Cortana, etc.). The application will constantly monitor blood glucose through passive collection and analysis of voice data. Guardian angel will use the information it gathers and analyzes to make dietary suggestions and lifestyle improvements.

Primary Feature 1

User: “Hey Guardian Angel, I just had breakfast. My blood sugar level is 85.” Guardian Angel: “Thank you, Wayne, I will store that information for you.”

Application will analyze a week worth of Users glucose data submitted by voice (Submitted after each meal for first week).

After collecting the base line data. Application will passively listen to voice samples of User. Application will monitor/store their blood glucose levels, making suggestions for correction when determined the glucose has surpassed a max or min threshold.

How it works Application will aggregate this data in two separate buckets. The First, is a collection of the User's Glucose information. The Second, is the actual voice data. Correlations will be made via statistical machine learning models between Users blood glucose and voice samples.

Note Machine learning models will become sufficiently more accurate with aggregate of all Users voice and glucose data.

Primary Feature 2

Guardian Angel: "Can I eat this ice cream now?" User: "Wayne, I am noticing elevated levels of sugar in your blood. Perhaps, you could try popcorn instead."


Guardian Angel: "Wayne, I am detecting low blood sugar in your voice sample. Perhaps, you would enjoy an apple." User: "Thanks for the suggestion!"

This feature can be understood more generally as a health/fitness monitoring solution. Application will utilize User specific data to make calculated food and lifestyle suggestions. Considering the current condition of our User, Guardian angel can provide meal suggestions or things to eat at any given time/place.

How it works An event based system allows our User data to be analyzed in real time with up to date information. Local restaurant information will be pulled from Google API's and device information will provide location awareness.

How we plan to build it

Application will live on a voice based consumer device (Amazon Alexa, Google Home).

Machine Learning Models will be built using Python and will be of type Multivariate regression.

Information will be stored and application hosted on the AWS platform.

Natural language processing will be handled by Amazon Lex (Underlying technology which powers Alexa).

Voice features will be extracted using the openSMILE toolkit.

APIs will be implemented primarily in Node.js.

3rd Part APIs include Google Places and Amazon Lex .


Tschoepe, Constanze & Duckhorn, Frank & Wolff, Matthias & Saeltzer, Gerhard. (2015). Estimating Blood Sugar from Voice Samples - A Preliminary Study. . 10.1109/CSCI.2015.184. []

Share this project: