MY ATHLETE
AI.
I connected my sports data to an AI system that I can query in natural language. Not to replace my coach, but to connect what Garmin records with all the context around it.
GARMIN RECORDS
MY TRAINING.
AI CONNECTS THE DOTS.
I am a technology enthusiast, also because it is my job. The starting point was simple: instead of opening different dashboards and looking at dozens of charts, can I ask a normal question directly to my data?
The goal is not to get another number. It is to add context: what I did yesterday, how I slept, how I am recovering, how my heart is responding, what the conditions were and how all of this compares with my history.
AI DOESN'T REPLACE
MY COACH.
IT WORKS ALONGSIDE HIM.
A prediction is not truth, and a model does not know everything a person knows. For me, AI is an additional analyst: it connects data, finds signals, proposes interpretations and lets me compare them with what actually happens.
ONE WORKOUT.
MORE CONTEXT.
After one of the last long sessions before Cervia, I did not just want to see distance, pace and heart rate. I wanted to understand how my body had responded considering everything that had happened before.
AI related the run to the very demanding bike session from the previous day, heart-rate response, environmental conditions, similar workouts in the past and where I was in the preparation. The interesting part was not how fast I had gone. It was understanding how I had responded to that load.
PREDICTED.
THEN RACED.
Before IRONMAN Cervia I asked AI to read my preparation and estimate each discipline. After the race I compared the prediction with what actually happened. Not to prove that AI “guesses right”, but to measure where the model reads my history well and where it gets things wrong.
Swim and run finished inside the predicted windows and transitions were very close to the estimate. The bike was slower. And it is precisely the gap, not just what matches, that makes the comparison useful.
THE RESULT IS
ONLY THE START.
After Cervia I was happy with the result. Then I asked a second question: beyond the finish time, what can you see when health, recovery and race data are combined too?
Value reported the night before the race, together with 6h40 of sleep and 98 minutes of deep sleep.
Resting heart rate reported on the eve of the race.
On race morning the device indicated a high recovery level.
The change immediately showed how much the day had drained the reserves estimated by the wearable.
Consumption reported in the analysis versus a device estimate of 4.706 L.
Two days later the watch still showed a long recovery, while HRV and sleep had already risen again.
THE FINISH TIME TELLS ME WHAT HAPPENED.
THE DATA HELPS ME ASK WHY.
CONTEXT,
NOT CERTAINTY.
AI can find relationships that would be difficult to see across many separate screens. But a correlation does not automatically become a cause, a prediction is not a promise and wearable values are not a diagnosis.
AI answers can also sound too certain. Part of the project is learning to distinguish measured data, interpretation and hypothesis.
That is also why the project is not intended to replace a coach or healthcare professional.
QUESTIONS,
NOT DASHBOARDS.
How is my fitness changing over the last few weeks?
What relationship is there between sleep, load and training quality?
With the current data, what performance is reasonable to expect?
Where did the prediction miss compared with the real race?
I WANT TO SHOW
HOW I BUILT IT.
This is also one of the themes I want to bring to MAV73 and Instagram: explaining how to build a similar system from your own data, using accessible tools and without necessarily subscribing to expensive services.
Not to copy my setup, but to understand how to collect the data, organise it and, above all, ask better questions.
PERSONAL DATA
NEEDS CONTEXT.
Athlete AI knows my personal history. The IRONMAN Data Project is building a second layer: the context of athletes, races, splits and rankings.
EXPLORE IRONMAN DATA PROJECT →