Built an AI coach for triathletes that reaches out to you - not the other way around

[fliellerjulian]

Hey everyone,

I’m Julian - AI Engineer, triathlete, and someone who got tired of being his own data analyst every morning.

Most of us paste screenshots into ChatGPT and hope for the best. athletedata is different: your coach already knows everything and messages you first.

7am. WHOOP recovery at 31%. Garmin body battery at 12. Your plan shows a hard threshold session. athletedata already swapped it for zone 2…and told you why.

After your ride, it breaks down power and heart rate drift against your zones, flags if aerobic base is slipping, and adjusts next week’s load. When your TSB tanks too far before race day, it catches it before it becomes a problem.

You don’t have to ask. It just knows.

Connects Garmin, WHOOP, Oura, Strava, Wahoo, Withings in under 2 minutes. Coaching via Telegram - no dashboard, no app to open. Built with triathletes in mind since the multi-sport data picture is genuinely complex.

Free trial at athletedata.health ; no strings attached.

Curious: for those of you who are self-coached, what’s the part of your morning routine that costs you the most mental energy?

Happy Training,
Julian

[RandMart]

@fliellerjulian

what’s the part of your morning routine that costs you the most mental energy?

Logging on for work - does that count?

[fliellerjulian]

Ha, maybe I should integrate Outlook so the coach can call in sick for you on low recovery days

[curtish26]

I am self coached. The part of my morning that cost the most mental energy is figuring out what to wear for the workout. My first workout this week it was 51 deg. F. The second workout it was 13 deg F. The 3rd workout it was 31 and snowing so every day I have to dress different for the conditions. I map out my training plan for an event 3-6 month out. So there isn’t too much to worry about in the morning. I up load all the workouts for the week to Training Peaks on Sunday. I can load them all on Monday of the following week (free version) to see the planned distance/time/TSS for the week. At the same time I can see the totals from the previous week to check if I am on target will the plan or not. I load the TSS for the previous week into my excel worksheet to calculate Fatigue/Fitness/Form/etc. (free TP version). It takes about 5minutes to review data and about 25 minutes to load workouts into the schedule for the week. So…30 minutes a week to tweak and plan the week’s workouts. I spend more time than that on the slow twitch forum a week. I do spend 8-12 hours writing out the overview for 3-6 month blocks though. I do that in between events when I am taking time off from training.

[adaykin]

How have you tested any of this out to confirm that it’s actually improving performance compared to the highest quality available alternative?

[fliellerjulian]

The outfit problem is real and honestly underrated as a cognitive load issue haha
Especially when temps swing 38 degrees in the same week (!)

I´ll add weather context to the daily coach briefing today - will keep you in the loop once its live in case you want to give it a try

[fliellerjulian]

Honest answer: we don’t have RCT data, and I’d be skeptical of anyone who claims they do for an AI coaching product right now.

What we can show is that self-coached athletes miss signals that are sitting in their own data - HRV dropping for 3 days while they’re still hitting hard sessions, pace drifting faster than prescribed across easy runs, volume spiking 40% without noticing. Not because they’re not paying attention, but because that data lives in 4-6 different apps that don’t talk to each other.

If you have a great coach who sees all of that, athletedata probably isn’t for you. If you’re self-coached and making decisions without the full picture, that’s the gap.

[fliellerjulian]

the weather integration is live now in case you want to give it a try: https://athletedata.health/

[trailblazery]

1. returning to www.athletedata.health does not keep you logged in. for people who don’t train with HR, your “app” is pretty useless. 24 hour training and ATL = 0. Great!

[fliellerjulian]

Hey, thanks for the blunt feedback!

both issues are fixed:

1. Logged-out homepage: if you’re signed in, athletedata.health now takes you straight to your dashboard. The session was always still valid (you could hit /dashboard directly), but the landing page wasn’t checking auth so it looked like you’d been kicked out.

2. ATL = 0 without HR: this was the real bug. Training load was being computed only from Strava’s suffer_score, which is HR-only - so a cyclist riding 14h/week with power but no strap got nothing. New behaviour:

• If you have a power meter and set your FTP, we compute proper power-based TSS from normalized power.
• If you train with HR and set your LTHR, we compute hrTSS.
• Runners can set threshold pace and we compute rTSS.
• Worst case (none of the above set), we estimate from duration so the chart isn’t empty - and the dashboard tells you it’s an estimate.

There’s now a small form right above the PMC chart on the analytics page where you can set FTP, LTHR, and threshold pace. Saving recomputes the chart immediately.

Would genuinely appreciate it if you took another look. And if you spot anything else that’s broken, please keep the feedback coming - it’s the most useful thing anyone can give me right now.

[fliellerjulian]

Hey everyone - another update from what shipped over the weekend:

What’s new:

• Shipped the Analytics dashboard (screenshot above): training volume, pace trends, aerobic efficiency, wellness & readiness trends, sport/zone distribution, all in one view

• Added a weather integration so the coach can factor in conditions when recommending what to wear or when to schedule your runs and rides throughout the week

• Requested additional Garmin endpoints to unlock richer, more actionable feedback

Coming next: TrainingPeaks integration is in progress.

The feedback loop has been incredibly valuable - I’m getting a steady stream of ideas via Telegram and email and building quickly based on what you all actually want.
If there’s a feature or integration you’d love to see, just reach out and I’ll make it happen.

To thank you guys for all the feedback so far, I set up code SLOWTWITCH that gives you a 30 day trial instead of 7.

Happy Training!

[curtish26]

@curtish26

fliellerjulian

ther integration is live now in case you want to give it a try: https://athletedata.health/

50+ deg. F = shorts & t-shirt

45-50 deg. F = shorts & long sleeve shirt

40-45 deg F. = running tights, long sleeve shirt

35-40 deg F. = running tights, short level shirt, running jacket, running gloves

30–40 deg. F. = running tights, long sleeve shirt, running jacket, running gloves, hat that covers ears

25-30 deg F. = running tights, warm-up pants, short sleeve shirt, long sleeve shirt, running jacket, running gloves, hat that covers ears

20-25 deg F. = running tights, warm-up pants, short sleeve shirt, long sleeve shirt, running jacket, running gloves, hat that covers ears, neck gator/buff

10-20 deg F. = running tights, warm-up pants, 2 x long sleeve shirts, running jacket, ski gloves, hat that covers ears, neck gator/buff

0 -10 deg F. = running tights, warm-up pants, 2 x long sleeve shirts, running jacket, ski gloves, hat that covers ears, neck gator/buff, two pairs of socks.

This is a lot of mental stress but having the right number of layers so you are not over or underdressed for a workout it totally worth it.

I will check out the AI platform. I am not too old to learn new tricks. IA is definitely a game changer for the speed and quality of training guidance.

[RandMart]

There should be a prompt ~30* that says “drive to gym, run on treadmill; do weights afterwards”

YMMV

[fliellerjulian]

That clothing guide is gold, saving this!
And yes, please do give it a try… would seriously love to hear what you think. Happy to help get your data connected if you run into anything.

[timbasile]

How much historical data is it pulling from Garmin connect or Strava?

[fliellerjulian]

Garmin is limited to 30days (their API limit). For Strava, it can pull all historical data

[islandergold]

I’d love the ability to accept or reject the proposed training plan update (before it makes the change). I don’t love waking up to a new plan and sometimes prefer to stick to my schedule in light of something in the data that AI may pickup.

great work overall

[fliellerjulian]

Thanks for the feedback! Its currently instructed to make changes proactively - but you can just tell it to ask for approval first and it‘ll follow your preference.

[Simon71]

Imagine going back a few decades and telling them that in 2026, knowing how to get dressed would be a stressful cognitive load issue requiring AI assistance!

[fliellerjulian]

lmaoooo