Built an integrated health dashboard tracking 30 years of medical data

As a veterinary doctor who has worked for almost 30 years integrating biosensor data in dairy cow health evaluation and early diagnosis. It was very frustrating not to be able to do that with my own health data: different doctors, different tests, different data formats, different scales, different measurement types. Everything goes into PDFs, into folders, and nobody can do anything with that information. Now add to that biosensor data like the Whoop and everything starts becoming impossible to decipher. The very first thing that I did during the course was to build three trackers for my daily habits:

1. One for sleep data taken automatically from the Whoop with an MCP connector directly to the Whoop data.

2. Another one to track all of my foods without having to go through the very cumbersome available apps out there because I very seldomly change what I eat. I just created a very simple one-page tracker with pills of my typical foods that I can just tap on and can change amounts that automatically update macros in the database.

3. The last tracker I built was one to substitute the journal that the Whoop app provides because it is too cumbersome, having to click multiple times on different behaviors to be able to enter the data. Again I just created a one-page app where all I have to do is select the pill that has the behavior and it all gets saved. I built guardrails such as "night meal" automatically switches "daylight meals" off nd viceversa, and other fail-safes, such as not allowing me to enter data in a future date, etc.

All three trackers save the data into a database in Supabase, along with every single blood test, every single body composition test, a DEXA scan, my DNA, and every single other medical test we have run, like cardiology fit test, etc. Then I created a dashboard that integrates data from every single source and plots it against each other. I access this dashboard on my iPad and it has multiple different graphs that vary by granularity depending on how long the period of representation is. I'm still working on some minor bugs but I've actually been able to do one of my last functional medicine consults with the app and the doctor was very impressed. We got to see things that Claude pointed out, when I created a skill to automatically update the dashboard any time that I put a new piece of information into the health dashboard folder. We have started to run statistical analyses to evaluate which behaviors or foods influence the outcomes. Pretty cool! 

The tools I used were Claude cowork, Supabase, GitHub, and Claudeflare to keep everything safe and locked behind a wall so that nobody can access my health data. What I learned from this project, which again was at the very beginning of the course, was that pretty much anything I could throw at AI, it would find a way to handle it. I have tweaked it in the last couple of months a little bit and the new tools are making things a lot easier. That's another learning, that once something is built it can still be improved later on as the tools improve. I am actually running every tool I've created so far through an audit process to see where it can be improved now that I have more data and better tools. I'm not going to lie, one of the learnings was how frustrating it could be to work with a tool that wants to be imaginative when I needed to be accurate and precise. Through many, many, many hours of iteration and tests and trials, I've managed to build an AI that works the way I like to work. I still feel sometimes as its secretary fixing some things here and there but the amount of hours that it has saved me looking at methodology has been amazing.