Sprint for the Buoys

“If one does not know to which port one is sailing, no wind is favorable.”

Lucius Annaeus Seneca

I’m in the sprint to the finish with my degree program while working full time. I’ve also promised myself I’m going to write *something* every week, even if it’s drivel. With the other commitments on my time, expect a lot of capstone and design thoughts.

I’m working on my capstone now, just fleshing out the design of my project. I’ve been given a fair bit of freedom in the subject, with the requirement that machine learning serves a key role. My initial idea involved building a system to collect baseline home health care data from patients with chronic medical conditions, and design a machine learning algorithm to measure the effectiveness of potential medical interventions. The system would collect heart rate, temperature and sleep quality data from a wearable like a FitBit, collate that with data from patient outcome surveys, blood pressure readings and other sources – either smart devices or manually entered, and finally match that whole dataset up with a second dataset of attempted interventions – things like specific physical therapy, adjusted drug doses, or other ideas posed by physicians, PTs and other care providers.

The idea is still near and dear to my heart. I hope some day a system like that can help people like my sister – people whom doctors are confused about how best to help and are casting about for quantifiable responses to experimental care changes. That said, the project quickly gets out of scope for a capstone. I hope I’ll have time to pick this idea apart over the next few years.

I’ve settled on a more limited scope for my current project. The system will take in a photo of a pill bottle lable, extract the prescription sig, then create medication reminders on the patient’s Google calendar. I’m hoping this can also be expanded a bit over the next few months. My initial plan is based around a Java desktop app, a Python back-end (using PyTorch or NLTK), and a Flask API sitting in the middle. If I keep these parts well-isolated, developing a mobile app client should be straight forward.

I’ll be covering more pieces of this project in excruciating detail over the next few weeks, but for now I’ll just chuck my first couple of design sketches here. I will also make no apologies for my horrendous handwriting: I’m far more comfortable typing.