Our Inspiration
Imagine the life of a Type 1 diabetic. Every three months, like clockwork, there's the inevitable visit to the doctor. It's a routine that's as predictable as it is necessary. Yet, despite the regularity, each visit feels like stepping into a realm of uncertainty. The doctor's office, usually a place of answers, becomes a theater of guesswork. The protagonist? My average glucose level over the last three months. The plot twist? A series of graphs and charts, sprawling across the doctor's screen like a city skyline, each peak and trough a story of its own. Then comes the moment, always expected yet never fully prepared for. The doctor, with a mix of concern and curiosity, points to a spike in the glucose graph. "Now what happened here?" he asks, eyes fixed on a sharp increase that occurred two weeks ago, on a Tuesday at 4 PM. The room falls silent. How am I supposed to remember that? This isn't just a question; it's a glimpse into the everyday reality of living with diabetes. A moment that should have been insignificant now becomes a mystery, a puzzle where the pieces don't seem to fit. And there I am, trying to sift through memories, searching for an explanation that seems just out of reach. The doctor, armed with years of experience yet hampered by the lack of detailed information, makes an educated guess. "Just try taking more insulin next time." It's well-intentioned advice, but it feels like a shot in the dark. And in the world of diabetes management, uncertainty is the last thing you need. This is where our story needs to change. This is where we come in.
The Solution
We will provide diabetics with a way to predict their glucose level based on the personalised data they provide to know when they should take insulin, as well as provide better data for their doctor to make more informed decisions. Our solution both allows for personalised suggestions to be generated for each user and minimises the risk of overshooting or undershooting when taking insulin so the amount and timing of the intake is based on science and data instead of the good old trial and error method.
How We Achieve This
GlucoBot takes a user input in form of natural language (you can just say you ate a slice of toast) and uses the data collected from the sensor over the past 24 hours to train a unique model for the patient and predict the glucose level in the next hour with over 80% accuracy. The suggestion is then filtered to see if the glucose level will warrant an injection of insulin and by how much which will then be fed back to the chat-bot to give the user the simple output of their predicted glucose level and insulin suggestion. Since every individual and metabolism us unique, it is crucial that the decision does not depend solely on a simple formula across everybody but depends on a Ridge Regression algorithm to provide individualised advice.
Log in or sign up for Devpost to join the conversation.