A Look into Athletica’s AI-Assisted HRV Monitoring with Andrea Zignoli

The imagination of many businesses that include your Oura Ring or your Whoop Strap or even your Garmin, the device that we use. There's so many platforms that are leveraging this concept of variability and how it can be used to give insight into how much stress is in your system.

Hello and welcome to the Athletes Compass podcast where we navigate training, fitness and health for everyday athletes. How do know if that feeling of a little bit of lack of motivation and fatigue is really fatigue or maybe just a little laziness or I don't really have it today? HRV or heart rate variability is one of the most effective tools to measure your overall stress and training readiness.

We discussed HRV back in episode 14 of the Athletes Compass podcast. But today we have a special guest, Andrea Zignoli, one of the architects of Athletica and its use of HRV. Andrea was back with us on episode 47 and 48, but Paul, could you start by talking about HRV and kind of giving us a background again of what we talked about and how it can be a tool for endurance athletes?

Yeah, for sure, Paul. So let's start back at the beginning a little bit. What is this heart rate variability thing? And it's right in the word itself. We actually listen to the word heart rate variability. We're actually looking at the variation beat to beat of that heart rate. You know, when we first look at, you know, the heart rate itself, we kind of think about it like a clock, you know, tick, tock, tick, tock, and it's just always consistent like a metronome.

Of course, now we know that that's not the case at all. If we really look closely at those beats, we find out that they're actually quite variable beat to beat. And the amount of variability gives us a lot of insight into our central nervous system and how stressed or calm we are.

The imagination of many businesses that include your Oura Ring or your Whoop Strap or even your Garmin, the device that we use. There's so many platforms that are leveraging this concept of variability and how it can be used to give insight into how much stress is in your system.

why is it important to look at that stress? And specifically when we look at central nervous system, we're talking about what's called the autonomic nervous system. So your automatic nervous system. And that's something you don't really intuitively kind of think about, right? It just kind of happens. And it can be, there are a lot of, if you're in a lot of stress, we say that you're in your sympathetic system and that's the fight or flight system, right? We use that system, that edge of the nervous system.

when we exercise. So it's really important when we're thinking about stuff we do in Athletica. Of course, you can't always exercise, you can't exercise all day. You've got to go to the other side of that balance area and that is the parasympathetic side. And that should be massively activated when you sleep, say for example, when you're resting, when you're recovering. So we're always sort of swinging back and forth between these two systems. And that's

That's what we want to make sure that we do that correctly. It's instrumental for our adaptation as athletes.

Now, specifically in the context of training, if you are doing, why this is important relates kind of back to something that we're also known for within Athletica is that our HIIT training, right? Our HIIT science. And we know from many studies that if you do HIIT training when you're stressed, when your heart rate variability is non-optimal,

you

you actually don't adapt very well to it. Studies are very clear with that. But if you do HIIT training where you're ready to receive that HIIT when you are more in balance, then you adapt very well.

So Andrea, you are one of the main architects behind the development of HRV in Athletica. You and Paul, with help from Dan Plews and Martin Busheit, wrote a paper about that, about AI-assisted HRV monitoring. Can you talk about this paper and what conclusions you came up with?

That paper came in response to a, let's say, of action from Martin in an editorial. And Martin Bushite, of course, was mentioning some concepts related to science.

2.0 and 3.0. And as soon as I read them, I was fascinated with the idea that we needed to refresh a definition for science, 2.0 even. We know that 1.0 was the definition of foundational concepts and 2.0 was referred to a

use of technology in sports science and coaching. And I said, do we need a new definition for 3.0? Apparently, we have new tools now and new understanding of what we can do with these tools. So we needed something new, apparently. And we started drafting this manuscript. And I soon realized that we were kind of repeating the same concept that

were already known by great scientists. We were standing on the shoulders of the giant, of course, and we were trying to write down what we knew about HLV and resting heart rate and the response that we can highlight to the training load. But that was not enough because we knew that that was something that we already knew and there was nothing

Thanks.

truly innovative in applying the science that we already have. Briefly, we know that if HRV or resting heart rate, they are deviating above or below a

the so-called normal range, we can raise a flag and say, there's something meaningful. There's something of note going on here. It might be that your HLV is rising above your baseline and going beyond your normal range, or is dropping below and...

Okay.

And the same goes for the resting heart rate. And I will let Paul then expand on the meaning behind these behaviors, of course. But our job in Athletica was more

removing the cognitive burden for the coaches and do whatever computation we need to do automatically on the platform.

We have been fascinated with technologies since the beginning of Athletica. the latest technologies related to large language models are not an exception. And we are fascinated from the power they have and the capabilities they have. So we try to explore that possibility.

of the principles behind the implementations of technology in Athletica is that

we would like to be sure to have control over the output that the logic is given to the users. So we need to have control over the output of something big as a large language model.

Amazing, mate. It's an awesome paper and we'll link to it in the show notes. And what I was hoping we might even do here is just kind of show people a little bit more of the concepts that are in Athletica. I'm kind of ad-libbing here, Paul, I hope you won't mind. But someone here on this call might recognize this profile. I'm not sure if that person is, want to put their hand up?

Ha ha ha!

Hello.

the

So Marjaana is, she's great. We've got her heart rate variability profile over, I guess we're just in here the last six weeks. We'll just look at that. But there's a couple key things that people, if you're a user of Athletica, you can look at some of these concepts that...

Andrea has been discussing here, but we've got a few key parts here. So you can see the band, that orange kind of band that's sitting in there. That's Marjaana's normal range. that's been calculated, correct me if I'm wrong, Andrea, but that's been calculated over kind of a 60 day rolling average of that variable. And that comes from Marjaana's Garmin watch when she sleeps with that overnight. And that feeds into the

to the algorithm here and we can see what's kind of normal for Marjaana. Now there's some, and then what we also have, the other one that's, sorry, each of these bars is also, that's every single night and the Garmin unit takes, I'm not sure if you know, Andrea, but I think it takes like an average of the lowest value of the evening. Do you know what it actually takes to get that lowest value? Like it's a nocturnal reading, right?

Yes, yes, it should be. It should be.

Yeah, so it gets the lowest value of the night. it uses a principle of consistency for reliability. And then that feeds in and it can kind of get an assessment of every time of day. Remember that circadian rhythm, so that lowest point of the night. And remember that low, high is kind of good and low is kind of bad, like we're generalizing. Low kind of means

there's a lot of stress, there's a lot of sympathetic in the system. High means usually there's a lot more parasympathetic activity that things are kind of like the system's rebounding. Case in point, we've got like a race right here that Marjaana just did. Marjaana just won her first ultra distance run that she's ever done and she was victorious. So congratulations Marjaana on that.

Wow, awesome.

but you can see her system is even like adapting right now. So this is very, it's almost like this, there's this super compensation principle that we often see within HRV. And we've got a textbook kind of example here for Marjaana. And Andrea's work here with large language models is kind of things are green in Marjaana's world. Notice the green color up here and it says, basically,

your seven day resting heart rate and seven day heart rate variability, baselines are stable and within the normal range. So this is the dark blue line that it's kind of following across here. so everything's copasetic right now for Marjaana, but we were actually in this, we were talking to Marjaana back here, I see is October.

28th, notice that the seven day rolling average was dipping below the normal range. And things were not as good back then, Marjaana There was a little bit of stress in life. I think your husband was away. So this was picking up things that weren't even part of Marjaana's training really. But she was telling me that there was kind of a lot of stress in life. Isn't that right, Marjaana?

Yeah, sleeps, mom's taxi duties, no help, cook supper whenever I could. Yeah, was a of extra stuff that had to be handled by myself. So yeah, it picked it up.

picks it up. So it's kind of cool. And yeah, I don't know, I just wanted to provide that as a little bit of a context thing for individuals. Now, and I'll just say like, this kind of doesn't exist anywhere else. I mean, if you've got the Marcos heart rate variability for training app, that's cool. You'll have a little bit of this insight there from that.

There's not too many others that are really going to now. I know what Andrea and his team are working on and he can expand on this, but they're trying to kind of link this back to, and this is mentioned in the paper, they're really working in the backend on actually integrating this into more, we say contextual factors in terms of like, well, now what do we do with this info? Can we make recommendations on the actual training that's coming up?

I think that it would be interesting to see when we start closing the feedback loop with the logic. There's still some answers that we are looking for because we, of course, we have a policy inside Atletica that we want never to fail prescribing a higher load, but we...

always be failing being too conservative because it's the safest approach and we want to find something that can work for sure for the average. And we have this internal tension between those who might be advocating for something more.

active in terms of logic. They want the logic to take control over the training plan. And those who advocate for the logic to be more passive and say, just raise a warning and leave the choice to the user. I'm more for those that would like to raise the warning and leave the choice to the user. But then some analysts can come up and say, I expected the AI to do that for me. So there's kind of these kind of tension and these kind

of choices that we need to make. But at the end of the day, what do we care about is having the data explained for the user the best we can without failing in doing that, making the data more accessible.

think we have both types of users. Just like Andrea is saying, we've got the user that comes in. They just want the AI to take control of it. They don't want to even think about it. And so we get that. But at the same time, we really feel that also, you as a user, coach, or athlete, you understand the context better than the AI can. Remember that. Never lose sight of that.

So what is probably the superior option is almost as we have things right now where we're gonna give you as much info as we can to help you out and make those decisions, but you need to take some control with your training, right? Like, again, back to that example with Marjaana, she's green and good to go, but at the same time, she's told me that she's got some...

some serious DOMs going on based on, you know, DOMs, delayed onset muscle syndrome, like she's got some good muscle soreness from that ultra that she just did. She's gotta be, you know, smart about what she does and how much she kind of does in this next little bit. And that's, you know, but she's kind of a more experienced user. She knows to use that common sense and she can take all the warning lights or not that Athletica gives her to help kind of make those decisions.

you

And that's with Paul and Marjaana who are experienced coaches, they can also help athletes make those calls. And I think to Andrea's point, we'll have sort of both options in the future, probably a box tick. Do you want the AI to take the full control? Do you wanna have a little bit more just advised sort of situation? So we recognize there's two types of users that are sort of out there on that.

I appreciate their conservative approach. And I also appreciate like, it must be so difficult to take one measurement like HRV and resting heart rate and then prescribe or change or decide to change or alter the future training. I can see the point of.

Mm-hmm.

somebody wanting that data to go in and change their next day's training. But I would almost argue that if you're just following the self-driving car, you're never going to learn to listen to your body signals and taking the whole picture, which is your life, your relationships, your work, all the stress that goes into the same bucket.

with your training, sleep, nutrition, everything. Like if you just look at AI program, whatever AI program you're using or cookie cutter program, you're not really getting in tuned with your own body and learning how it works. And you're gonna crash it at some point for sure. I've done that when I first started with triathlon.

And I was using HRV measurements, but I didn't really put that in the context. So I was just like ignoring, ignore, I'm doing this anyway. This is what a training plan is asking me to do. So I can appreciate the difficulty to, you know, look at the next step

You know, when I was just kind of thinking, when I let off with my intro, I didn't really speak about, you know, the big picture of why it's important to actually even look at this heart rate variability. And I missed a big piece that I just want to reiterate here. So when you get any training plan, you can be like, there can be two athletes that look

Very, very similar. We've spoken to this before in former podcasts, but imagine like these two identical looking athletes that can do this training program, right? And any training program on the internet, we can do that. But those two identical looking athletes, remember that they are going to respond differently to that same program. This is where heart rate variability comes in. It's one of many, what we call load response

markers, right? Heart rate's the other one, how you feel, but all these emotional things, these stresses that we all have in our life that we're dealing with, a bad sleep, poor nutrition, alcohol, all these things are gonna affect us to have a different response than our identical looking brother or sister. So keep that in mind, heart rate variability.

and has been discovered as a sensitive marker. But there's other things too that we have to look at that we are in the context of Athletica, such as if you're entering in your subjective comments, right? Like that we use something called semantic analysis to actually assess how that comment kind of relates. And a coach can do that too. Then there's also the rating perceived exertion and the feel. So the challenge that

Andrea and his backend team have are capturing more and more of those contextual factors in alongside the heart rate variability to help us make those decisions or just recommendations, whatever they might be.

is nice discussion. If you go on the forum, there's one of my favorite discussions about the destirability of the system, how fast you can change the next training sessions. If you write the worst feeling possible in the text box or you write RPE 10, I'm exhausted, please help me.

Ha ha ha

you

The car won't steal that fast anyway. It's a choice that we made to, again, we designed the system to cope with the average athlete, of course, and it might deviate from what one might expect or need. we have this approach because we're talking about type one and two errors like...

We are diagnosed, for example, this is something in diagnostic that it's interesting to talk about. If you send someone to the doctor when it's not needed, it's different not to send them to the doctor when it's needed. So it's different kind of error you're making. And again, we want to play it safe at this point in just highlighting that some changes to the training plan are needed. And we wanted to highlight that.

the logic might be acting on those changes. users will always remember that there is the option in the workout wizard that is always available just in case they wanted to take the opportunity to change something in their training plan. And of course, they can move session around. Again, some might say that the AI coaches should do that in my place, and I shouldn't be doing that.

We cannot drive a self-driving car 10 meters from the wall at 100 kilometers per hour and then say, okay, save me. And we cannot just expect it to break the law of physics, the car would crash anyway. So we should be very...

why is unaware of what we can ask to a self-driving car to do and what they cannot do. But it's not because the technology is not there, it's just that it's something that defies the law of physics pretty much because the car has an inertia and of course your training preparation and your fitness also has an inertia so you cannot stop things like from a day to another.

anyone listening to this, just to give them some clarity on when it's appropriate to take some action, and what sort of action you can take. So if you're, would just kind of say, and I'll, I'm just putting my coaching hat on. So, and I'm just imagining I'm coaching and advising Marjaana, which I do from time to time. And you know, if I would just kind of say that, you know, if we are seeing.

things on Athletica that, first of all, is the feeling, you know, is the motivation to train a little bit low. That's the first sort of red flag that I would say, okay, let's have a think about it. Is there anything else? Is there any further info that we can get? Well, are there any warning lights on Athletica in terms of, and actually we're looking at the load and you'll see these on your Athletica plan, right? Like just on the training plan itself, it's up on the left.

Good.

left-hand corner, there'd be just these kind of warnings. that's, there's usually looking at the external training loads, like how much pace or power that you've kind of been pushing over durations. You know, have you been really pushing it a little bit? And, you know, is there a warning there? That's a kind of a second one. And then third, is there a heart rate variability warning as well, right? Is that one kind of going down? If you're getting three red flags like that, know, your motivation's low, you know,

you

external load warning is there and your HRV is in the tank, you know, seven day relative to 60 day. That's a pretty clear message that you want to do something different. don't, maybe you should be doing something that's, you know, an easy aerobic session or just resting. So, and I would just kind of say, like that's good advice for coaches. That's good advice for athletes that are self-coaching.

Thank

So one of the things that I liked in the paper was that Andrea was, I'm assuming that you wrote this, but that AI should be viewed as a co-pilot, not as a replacement. And that seems like the kind of the theme of what Paul and Marjaana were just saying and you too as well, Andrea, is that HRV, resting heart rate, are tools that we can use to monitor how we feel.

but we still need to make a decision about whether this training is best for us. Can you talk more about the role of the athlete?

What should the athlete do in terms of looking at all the data? Paul, had a list of things. What can we do as everyday athletes to look at our data?

Mm-hmm.

I think that we wrote that in the paper not just because we wanted to save a job to anyone, clearly. It's just like an objective reflection on what is best to do.

for both sides. I'm not talking for the interests of the human coaches or for the AI coaches. They have their own rights, so I don't want to talk bad about them. I'm just saying, and I recently gave a short talk for a class at the UCI course for advanced coaches. And in my last slides, what was exactly like

that was a quote taken from the famous movie, The Matrix, where before the end of the movie, the AI agent said to Neil, never send a human to do a machine job. And in my slides, I was saying also, never send a machine to do a man job. So we know both sides what they can do best.

So we have like machines, have a computation, we have interpolation.

and we have memorization. In humans, we have extrapolation, so generalization, the ability to generalize. And we have the emotional intelligence, of course. We have empathy. So it's not only convenient because we can save our jobs to leverage AI only in what they're best at, but it's also the right

engineering choice, let's say, the most practical, the most useful and the most effective.

So we don't want to use the wrong tool at the wrong time, we want to use the right tool at the right

So this is my point of view.

Perfect. Yeah, so well said, Andrea.

One quick question, is this limited to just Garmin devices for now?

Yeah, something maybe will change in the future with additional integrations, but they're not in the pipeline yet. that's the beginning, and we are kind of testing it out. We are seeing what we can do with it. So for the moment, it's gamming only, unfortunately, I would say.

of course.

So here are my takeaways from this episode. They're short. Number one, heart rate variability and resting heart rate can be effective tools to help us monitor our training readiness. Number two, AI like Athletica can provide a great way to crunch all the data and provide actionable guidelines to help training progress and explain the context. And number three, I am just gonna read right from the paper.

because this is so well said. This is the last note. While AI can efficiently process and retrieve vast amounts of information, human expertise remains essential for contextualizing that data, interpreting subjective inputs, and making nuanced training decisions. And I think that is kind of the whole point of what we've been talking about today. That's all for this week.

Join us next week on the Athletes Compass podcast. Ask your training questions in the comments or on our social media. If you have enjoyed this episode, we'd appreciate it if you would take a moment to give us a five-star review and a follow. We'd appreciate it as well if you could share this episode with one person, just one. That would be helpful. For more information or to schedule a consultation with Paul, Marjaana or myself, please check the links in the show notes. For

Andrea Zignoli, Marjaana Rakai and Dr. Paul Laursen I am Paul Warloski and this has been the Athletes Compass Podcast. Thank you so much for listening.