Not AI coach for triathlon and running since AI does to much funky stuff - I built RaceMind because life kept breaking my training plan

AI Tools
1

Hey everyone,

I’m a father of two, train for marathons and Ironman-distance triathlons, and like many age-group athletes I have a fairly normal problem: I don’t have unlimited time, my weeks rarely go exactly to plan, and a private coach is hard to justify financially.

Some weeks a kid is sick. Some weeks I’m sick because a kid was sick. Sometimes work runs late, sleep is bad, or the planned long session just does not fit anymore.

I tried a lot of training apps and plans, but I kept running into the same issue: most plans work well as long as life behaves or are just not good UX. The moment a few sessions are missed, the plan either becomes unrealistic, creates a “catch-up” spiral, or leaves me guessing what to do next.

So I started building RaceMind.

The idea is simple:

A training coach for marathon and triathlon that adapts when real life gets in the way — without blindly pushing you into more load.

Why the plan engine is not AI-generated

This is probably the most important design decision.

RaceMind does use AI for coaching, explanations, chat, analysis, and helping you understand what is going on in your training.

But the actual training plan generation is deliberately not generated by an LLM.

I tested that approach early on, and for me it was either:

  • too unreliable when it was fast, or
  • too slow and over-engineered when I tried to make it reliable enough.

For training plans, especially when fatigue, progression, injuries and missed sessions are involved, I wanted something more deterministic.

So the plan engine is rule-based and built around clear constraints:

  • controlled weekly load progression
  • no aggressive catch-up logic after missed workouts
  • conservative return-to-training after illness or injury
  • phase-based structure: base, build, peak, taper
  • sport-specific logic for running, cycling, swimming, strength and mobility
  • availability-based weekly planning
  • adaptations when the week no longer matches the original plan

The AI sits on top of that system as the coach interface — explaining decisions, answering questions, helping you adjust the plan, and giving context after workouts.

But the safety-critical part of “what should be scheduled next?” is not just an LLM making things up.

What RaceMind does

RaceMind is built for runners and triathletes preparing for:

  • half marathon
  • marathon
  • middle-distance triathlon / 70.3
  • long-distance triathlon / Ironman
  • Ultra running (in preparation, happy to hear feedback for it)

Some of the main features:

  • adaptive weekly training plans
  • automatic replanning when sessions are missed or moved
  • coach chat that can actually change the plan, not just talk about it
  • post-workout analysis in plain language
  • readiness guidance based on recent load, sleep/recovery data where available
  • conservative comeback logic after sickness or injury
  • race-day pacing and nutrition guidance
  • structured workouts that can be sent to your device
  • goal-time forecasting that updates as more data comes in

The main problem I wanted to solve was not “give me more data”.

Intervals.icu already does a great job with data.

The problem I wanted to solve was:

What should I actually do now, given my goal, my current fitness, my fatigue, and the fact that my week just changed again?

Why Intervals.icu matters here

I really like Intervals.icu.

Honestly, without Intervals.icu I probably would not have started building RaceMind in this form. It gave me a much better foundation for understanding training load, history and athlete context than most platforms.

That is also why I would love feedback from people here specifically. This community understands training data, but also knows where data alone is not enough.

Free for Intervals.icu users

As a thank you to this community, Intervals.icu users can test RaceMind free for 3 months.

Use this link/code: INTERVALSICU

No pressure to continue after that. I’m mainly looking for honest feedback from people who actually understand structured endurance training.

Website is here:

What I’d love feedback on

I’d be especially interested in your thoughts on:

  1. Would you trust a rule-based plan engine more than an AI-generated plan?
  2. What safeguards would you expect before an app changes your training plan?
  3. What should happen after missed workouts: skip, replace, reduce, or replan?
  4. How conservative should a return-to-training flow be after illness or injury?
  5. I’m also considering adding ultra running support. To me, that seems different enough from marathon planning that it should not just be “marathon plan, but longer”: elevation, durability, back-to-back long runs, fueling, hiking/walking strategy, and more conservative load progression probably matter much more. Would people here find that useful, and what would you expect from a good ultra plan?

Happy to answer questions, and also happy to hear criticism. I’m still actively building this and would rather learn from experienced users early than continue in the wrong direction.

Thanks,
Dominik

4 Likes
2

Lot’s of AI tools here - at least here it seemed like Claude did a good job.

3

Thanks, Bruce, but even with Claude, you need to know your way around programming :wink:

4

Interesting approach. I like the idea of not letting an LLM generate the actual plan.

Can you explain a bit more how RaceMind decides what to do when a workout is missed?

For example, if I miss a threshold run on Tuesday but still have an endurance session and a long run later in the week, does the app usually skip the missed session, move it, reduce it, or replan the whole week?

I’m asking because “catching up” missed intensity is one of the things I would be most worried about in an adaptive plan, especially for running injury risk.

5

Connection with intervals doesn’t work, AI doesn’t respond and plan is artificial and strange.

6

Hi,
I just had a testrun with 42 users, 10 of them with intervals, and no one had issues - also yes, the chat does for 1 in 300 occasions (I just checked in the db) get an error, but there is a fallback and if this also does not work, I do have error logging, same actually with Intervals connection. I just wrote you a private message, so you can give more details, since I also don’t see any error logs (could of course also be my bad implementation, but so far, they worked reliable).

For the plan you mentioned, that it is artificial, also here check my DM, since the plan generation was developed with two sport scientists and was found to be quiet robust in last testing round. It is also as I wrote not by Gemini but by an algorithm, since AI is too slow/unreliable.

So happy to resolve that, was just a bit puzzled by your comment👌

7
  1. It didn’t pull any data from Intervals, probably only FTP. There was no training history or the calendar of planned workouts that I have entered. So basically nothing important was imported.
  2. Because of that, the AI couldn’t tell me what my actual training load/fatigue status was and kept saying I had 100% readiness, even though that day I had swimming and a VO2 max run.
  3. The plan somehow generated automatically, even though I didn’t even approve it — I just backed out halfway through :slight_smile:
    4.The plan didn’t take into account my current training phase or the goals I had entered. It started me at 4h/week when I’m currently doing 9h weekly, and it would have completely ruined my build phase by also throwing in strength training sessions.

The idea is solid, but something isn’t working. I used AI Endurance for about six months before, but eventually quit because I’m a father of two, I work full-time, and I started feeling guilty every time I skipped a workout — which only led to frustration.

In the end, I train whenever I have time, and the only thing I really need is a decision. A clear answer to: “What should I do today?”

That means the AI should analyze all my weaknesses and undertrained areas from the past, take into account my current fatigue levels, the time I actually have available, whether I still need to do grocery shopping today or drive my kid to guitar lessons :slight_smile: — plus my race goal, current training phase, the weather outside, is swimming pool available, what’s on Eurosport, and even suggest music that matches the training load and my mood.

But right now, only my brain can do all that.

Good luck, fingers crossed :slight_smile:

8

Hi, thanks, now with this I can work better :slight_smile:

  1. the app uses your history only to get your current sport data, since you had a run event as goal, it only looked at your run data, it does not consider any planned workouts or goals you set in intervals, it really is just for getting your current fitness level for the event you selected - here is also where the 4h/week start was coming from, since the app does ask you for your current training time and it also checks your history (if available) - it takes both of them in consideration for planning the start duration, since if you are currently training 3h but you want to train 10h, it should not start with 10, since the risk of injury is too high
  2. when checking for your current readiness, it takes your hrv, acwr and sleep, since the sessions you had, were not done, when you already used the app, it accounts them as history, and therefore not considers them, this is indeed an edge case I will fix, but it can only occur on first training day, so on second day, it knows all trainings you did, since you already started with the app.
  3. plan generation is part of the onboarding, when you stop it in the process, it saves, the last plan, which was then the not approved one by you

I can totally relate to what you are talking about, since I am also a farther of two and have somehow the same problem - that is exactly why I did create the app - since if you don’t finish the plannend sessions, it suggest to adjust the plan - and, what was at least important to me, knows, what you current training load is and does not plan the next weeks according to the old plan, since this leads to too big of load increase, what again increases the likelihood of injury- what’s on Eurosport I have not added yet, but at least if you have a pool available, weather, race goal and current phase it considers :slight_smile:

9

Hi, sorry for responding that late, was on a trip with the family.

When you miss a workout it depends, when this happens, so if it is early in the week, it could get rescheduled on another day depending on your availability. If there is no slot available, the workout gets either skipped or replaces another one if it is a key workout. It also depends on the other workouts you have during the week, so for example that you don’t have two VO2 max sessions right after each other.
Injury prevention is something I tried to focus on, since I also had some issues with injuries due to overtraining/wrong training.

10

Ironman athlete here. I’m almost in tapering phase. Does it actually make sense to test the AI? The first years I used online platform training plans. This is my first year/season where I wrote my own training plan.

My own plan worked well so far. My main focus also is on injury management as others also stated. I think the biggest risk is overtraining.

I’m female, 55 years old.

11

So yes it does, just not for following the plan proposed by the app, since I would not change plans that short before an event - you can use it then more for feedback on your workouts.
Also I’m implementing currently a structured off season mode, there you then could also follow the plan - will write here, when this is ready :slight_smile:

12

Hey @asa14 very interesting approach! We have with https://youb.app exactly the same approach - we started a year ago and to be honest it seemed easier than it’s really be to combine llm as communication layer and workout logic from a human coach. We needed a whole cross functional team to have the expertise from sport science to the right UX Design - would be very interesting how you solve the gap between sport science and LLM?

13

That is exactly why in the end we decided to have the whole planning not done by an LLM - since testing with athletes and doctors showed, that it does overall provide good results, but not always, and this not always was too much, since injuries could follow from this.

14

With the newest update, there is now also an off season mode, which let’s you choose between maintain and improve speed - go check it out​:ok_hand:

15

For all parents, you now have an option to add a buggy or stroller to your workout, which then is considered in the analysis, so no ef drop anymore, when carrying 20kg of kids in the buggy with your bike or while running​:rolled_up_newspaper:

16

helo..i just installed the app. It looks cool!! How do I get the free 3 months??

17

you need to create an account and then you can redeem the code, got a dm :wink:

1 Like
18

I did.. and I see the free 8 days trial and a reedem a friend code

19

Ok…I did it

1 Like
20

So since last update with stroller und buggy support, we also added now vacation mode with training camp option.

Also we added our own power estimater for none powermeter rides with the bike.

Go check it out and let me know if you have an questions​:ok_hand: