I didn't set out to build a product. I was just trying to get better coaching for myself.
It Started with data downloaded from garmin
I train kickboxing. I wear a Garmin. Every session, every night of sleep, every morning HRV reading — it's all there in the data. But the data just sits in Garmin Connect, mostly ignored. I'd glance at my sleep score or check my max HR after a hard session, but I never really used it.
One day I exported a few weeks of training data and pasted it into an LLM. I asked a simple question: "Based on this data, am I overtraining?"
The answer stopped me in my tracks.
It didn't just say yes or no. It walked through my HRV trend over the past ten days, pointed out that my resting heart rate had crept up by 4bpm across the week, noted that my last three sessions all had recovery HR drifting higher through the intervals, and connected that to two nights of poor deep sleep. It told me I was accumulating fatigue and that if I had a competition coming up, I needed to back off now or I'd peak too early.
That was from a single paste of raw data.
Then I Kept Going
I started loading more data. Sleep from the previous night. That morning's HRV. My session data after training. I'd chat back and forth — asking about intensity, what to focus on, whether my interval performance was improving or declining.
And it kept getting better. With more context, the advice got sharper. It noticed patterns I hadn't seen — that my best sessions always followed two rest days and a night with 80+ minutes of deep sleep. That my HRV was consistently lower on Mondays, probably because I was training hard on Saturdays and not recovering over the weekend. That my interval max HRs were actually more consistent when I warmed up longer, even though I felt like I was wasting time.
This wasn't generic fitness advice. This was a coach who knew my data, remembered our previous conversations, and was genuinely tracking my progress over time.
The Tipping Point
After a few weeks of this daily routine — paste data, chat with an LLM, get coaching — I realised something. I was getting better insight from these conversations than from anything else in my training. Better than apps. Better than most human coaches, honestly, because it had perfect recall of every session, every sleep metric, every trend.
But the workflow was painful. Export files. Copy. Paste. Format. Re-explain context from yesterday. Every morning, the same ritual just to get a briefing.
That's when it clicked: this needed to be an app. Not because the world needs another fitness app, but because the coaching quality was genuinely too good to leave trapped in a chat window. If the data pipeline was automated and the context was persistent, this would be something properly useful.
What Fight Coach AI Actually Does
So I built it. The app pulls your training and sleep data automatically, builds a rolling picture of your fitness, recovery, and readiness, and gives you a daily AI coaching briefing that's grounded in your actual numbers.
It knows your HR zones. It tracks your HRV trends. It remembers what you talked about last session. It adjusts advice based on how close you are to competition. When it tells you to keep your heart rate in Zone 2 for recovery, it means your Zone 2 — not a generic guess.
On top of the coaching, I've been building out on-device strike detection using pose estimation and ML classification. Point your phone at yourself, hit the bag, and the app counts and classifies your strikes in real time. Combine that with the AI coaching and you've got something that watches your training from both sides — the biometrics and the technique.
Early Days
The app is still in early access. I'm using it daily for my own training, and the coaching keeps getting better as I feed it more data and refine the context. If you're a combat sports athlete who trains with a smartwatch and wants better feedback on your training, join the waitlist.
This started because I wanted a better coach for myself. I think it can be that for other people too.