What it does
Web app which will have two features. First is a classifier such that provided a cropped image of a blemish, we pass the image into a classifier and return what type of blemish it is. It may be a bug bite, rash, etc. Second feature performs sentimental analysis on medicinal reviews and rates medicines depending on their sentimental value. All of this we are deploying onto AWS SageMaker through docker.
How I built it
Used Ham10000 dataset and applied style transfer learning from resnet to learn the boundary decision for skin lesions with 85%+ accuracy. Using another insurance dataset to perform nlp analysis on.
Challenges I ran into
Getting SageMaker and docker to cooperate was the main issue of this project aside from its ambitious scope.
Accomplishments that I'm proud of
Classifier actually has a 85%+ accuracy rating after training for only a short while.
What I learned
SageMaker is really cool but very hard to work with.
What's next for MedESkin
Wrapping up SageMaker and finishing the second feature, along with polishing the front and backend connections.