What it is
A personal fitness app for one person, me. Built around a question I kept coming back to: given everything I know about my body right now, what should I train tomorrow?
It pulls together blood test results, Garmin watch data, and the goals I'm working on, and turns that into training plans I can run from. The plan lands on the watch so I can do the workout without needing my phone. There's a daily-health dashboard for the bits that aren't a workout but still matter.
Why I built it
The off-the-shelf training apps assume one input. Strava knows your runs. Garmin Connect knows your watch data. MyFitnessPal knows your meals. None of them are good at "here's my blood work showing I'm slightly anaemic, my HRV's been low all week, I'm trying to be lighter and stronger by August, what's the right session today?"
That kind of synthesis is exactly what AI is good at, and exactly what no current consumer tool does end-to-end.
So I built one for myself.
The interesting parts
Multi-source ingestion
Blood test results (manually entered for now), Garmin data via an unofficial library, training log, food log, sleep, HRV trends, and the goal I'm chasing. Each of those is easy on its own and the value is in stitching them together.
Personalised programming
The model takes all of that and writes the next session. Not in the generic-AI-coach way that picks a workout from a library and renames it to feel personal. More like: look at the body in front of you, look at where it's been and where it's going, and choose accordingly.
Garmin push, the hard part
The plan needs to land on the watch. I'm not a Garmin developer in the official sense (which costs money and time I don't have), so the route is through a sideload library that pretends to be Garmin Connect from a standard user account. Works, but it's the most fragile bit of the system and what I'm refining now.
If I can't get structured workout push working cleanly, the worst case is that the watch becomes a read-only input and the workout-of-the-day lives on my phone instead. Less elegant, still useful.
Status
In personal use, with rough edges. Not designed to share. The case study is here because the architecture (synthesising structured medical, wearable, and self-reported data into a single daily recommendation) is the bit that has legs beyond just me.