19A – Idea Napkin No. 2

1) I am Fractal, a Maker and lover of all things innovative. I believe in the power of machine learning and how much it can speed up the workflow and make our lives easier by taking out the grunt work in important but exceedingly repetitive jobs like mine. I work in data entry. I read prescriptions as they come and I type them up. I belive a robot can and should take my job. My job would then become double-checking if the robot analyzed the image correctly, thereby making my life a whole lot easier, and making the data entry process a whole lot quicker. 2) I'm offering an artificial intelligence machine learning program that is capable of detecting and handling handwritten text, especially for that of data entry. 3) My ideal customer would be corporate data entry offices, giving them the opportunity to implement an AI that automatically processes and data-enters prescriptions. With literally millions of databased prescriptions on file, there's practically an endless stream of training data for an algorithm to learn how to write them. The technology of reading doctor chickenscratch can then be deployed to students to digitize their hurried notes. 4) They should care because it's important for them to be able to enter in the prescriptions as quick and accurate as possible, and the average error rate is around 2-3% for my department as a whole. When you consider the in-office pharmacists' error rates, and that of the in-store pharmacists, it's entirely possible for a patient to get a wrong drug. I once was mistakenly dispensed the wrong strength of a drug simply because I had been prescribed it before and I suppose they mistook it for the correct strength. In my case this wasn't a big deal, but imagine if a patient meant to recieve a 5mg strength of a drug but was given 50mg instead! This could potentially be fatal for the patient. The AI would simply act as another step in the error checking, not just as a speed increase. 5) What sets me apart from others is that I already work in the field, and I'm proficient in machine learning technology. Few people work in both healthcare and in robotics or machine learning, which is probably why we don't yet have a real Baymax. With all the training data, we could also greatly improve machine learning technology while giving to the healthcare community--a mutual symbiosis! I believe all of these elements fit together very well. The only potential issue is that without the training data being manually outlined in the image, the learning algorithm will have a much harder time figuring out what to look for in an image. But with only a handful of outlined training data and supervised learning to correct the AI, it's possible it could be trained efficiently. As an oddly specific idea, there isn't much to be out of line with other things. As far as the "feedback memo" I am unable to write this part as I did not recieve any comments.