Cynder and a new direction for me.

The Cynder project was a passion project of mine from the very start. When I was born I was born with a rare disease known as Non-Cystic fibrosis. Sadly for me and 50% of kids diagnosed with the disease it is usually quite late until symptoms show up. Furthermore if symptoms do show up it would take a skilled physician or X-ray technician to notice the slight difference that proves the presence of Cystic-Fibrosis, Non-Cystic Fibrosis, or Bronchiectasis. This leads to many pediatric patients like me not getting correctly diagnosed. With this story in mind I thought to myself of how I could apply myself to create a solution. Having a skillset in data science, and AI it did not take me long to get random ideas. I thought about creating a AI system that analyzes DNA but to utilize that technology you would need genetic testing. I then thought about creating an app to look for family lineage of the disease. Sadly an app of that sort would be hit or miss. As such I went with my final idea, an idea so madcap that I thought that it was absolutely impossible. The plan was to make an AI training and loading system/configuration to train a custom image classification model. This odea seemed crazy to me as a AI engineer due to the lack of data in the field. Furthermore training a functional AI on a home computer, on the CPU is the equivalent of cooking toast in the microwave, it will surely turn out badly and may result in fire. Well with those truths in mind I began my coding, I worked on creating a system for training and spent hours getting the hyperparameters correct. Then I hand collected data on Cystic Fibrosis from non-copyrighted google images. I trained the AI program one, exported its model .CH5 file. Then I put it into my loading script, which tested the model by inputting a validation image and then printing its classification, in this case the presence of pneumonia or CF. After attempting to validate it against other lung diseases I proved my model to have somewhere around 96.765% accuracy. This was amazing to me, not only had I setup the model correctly, but I had achieved a great accuracy. It was at that moment that I attempted to see the limitations of my trainer script. As such I tried changing the data to instead by either Hookworm or Whipworm eggs. I ran the trainer and in around thirty seconds I had a highly accurate detection model for Whipworm and Hookworm eggs. This not only proved that my model had worked, but that my trainer had as well. It was at that point that I decided that the public would benefit off of my creation, as such I released it for free on Github, and dubbed it the Cynder project. (https://github.com/Arctic-framework/Cynder) The Cynder project is still in Beta but getting stronger every day, and personally I am excited to see where it goes, and what secrets it holds, but for now it will be identifying disease for free, and allowing new coming AI coders a chance at training there own models.