Latest / Healthier: Building Health & Fitness Businesses / How AI helps me get fitter every day (Strava, Garmin and Claude) | James McAulay | Healthier Ep 3
Transcript
- speaker-0: I did a Randox blood work test in November that came back with this huge 65 page PDF, like the most information I think I've ever seen on my own body. Kick off with a deep research task to summarize something you just heard about and then have this agent just... completely personalise it to your level. With any sport, but particularly with endurance sport, it's tempting for people to hear some influencer talking about running 100k a week and then just injure themselves because they've gone from like not to 100 far too quickly. And this system would immediately push back on that and say, look, I know what you've been up to over the last like three months of your training. You're not ready to run 100k every week. You're not even ready to run like 50, to be honest. Let's build this up so that you don't injure yourself. speaker-1: Hello and welcome back to Health Stacks, the show where we have experts, practitioners and enthusiasts come on and share their amazing AI and technology workflows for health and fitness today. I'm extremely excited to be joined by my good friend James, who is going to show us some amazing things as to how he uses AI for health and fitness. And think maybe a bit of a focus on fitness. So James, tell our viewers what they can expect if they stick around with us and what they're going to learn from today. And then we'll hear a little bit more about you and we'll pass over to you to get started. speaker-0: Yeah, absolutely. So I have done a couple of Ironmans and I'm training for a big triathlon this summer. And I've been setting up an AI system that helps me analyze workouts, analyze my wearable data from my Garmin watch, sets me gym sessions based on my triathlon specific goals, critiques my gym sessions after I've done them, and does all of this inside the cloud desktop app. using MCP servers. So yeah, in this episode, just, I'll show you exactly how it works. Set it up yourself. speaker-1: So essentially, if people have got like a long-term fitness goal and they've got some inputs that can be used, like their wearable data or their workout data, they might be able to copy what you're doing to get themselves a really, really high standard coach to help guide them along the way and hit that PB, like hit the goal that they want to hit. speaker-0: Absolutely. mean, I think if anyone's doing anything endurance related, they have a Strava account and Strava MCP is one of the connections I'll show you. So at a minimum, you can just hook up Strava to Claude or chat GPT and start talking with your data, like with a coach. I will say that like, we'll get into this later. This is not a replacement for a human coach and it's very effective for my own data analysis, but it's not good at like telling me off when I miss sessions. So we'll talk about like the sort of value of both, but yes, anyone can set this up at the end of this episode. speaker-1: Amazing. And before we get into it, what were you doing before you had this set up? So like, I know you've been taking this stuff very, very seriously. You've got a bunch of accomplishments, right? So you were, even though you've gone and set something up that's pretty cool from a technical point of view, like I assume you had a setup before that was doing okay. And like, were the, where were the gaps that you wanted to solve with this new tech setup that you've got? speaker-0: Yeah. So I am like an endurance sport nerd. And I have been since I discovered an app called Endomondo when I was about 17. So you wear a GPS watch or use your phone to track a run and you get loads of data out the other side. And for the last 15 years, I've been obsessively looking at my data from a session and then optimizing my training plan or my workouts to hit a goal. Um, in 2019, I started following like a structured training plan for the first time in my life. So like a lot of people, I wanted a 5k PB. I wanted to run under 20 minutes. Um, and every time I went out for a run, I just tried to hit that PB. Didn't work. Um, but in 2019, I joined the triathlon club and I started following, you know, structured workouts and, uh, proper, uh, proper training plans. Um, so using Strava. I was using spreadsheets, like I was using a lot of spreadsheets. I discovered a great tool called trading peaks, which is how me and my triathlon coach were sort of setting our workouts, reflecting on our workouts. I've been using a Garmin for about eight or nine years. I used a whip from about 2020 to 2023 and then unsubscribed. speaker-1: I famously gave that up. But we can claim us that later. speaker-0: Yeah, we can get into that later as well. But yeah, I've been, I guess tracking everything with Garmin for years, tracking everything with Strava for years, and then cobbling together my own spreadsheets that I can show you if you want to. speaker-1: And now it's evolved. Let's dig in. Let's have a look at the setup as it is now and what it's doing for you. Maybe you can talk us through it. speaker-0: We're in Claude and I've set up this Claude project, which is one, like a way of keeping everything organized, but two, a way of giving an AI agent specific instructions and specific data to help you achieve an outcome or a workflow. So I've said to this agent, you're helping me get back into the best shape of my life. I've done two Ironmans before. And I want to perform well at a T 100, 100 kilometer race in the summer. Now this is just, think like the description of the project. ⁓ so this isn't the specific instruction in here. I have been gradually evolving the system prompt for this, ⁓ agent. So I've said, this is a mandatory data protocol when we are talking about. workouts for training data use Garmin MCP, heavy MCP, Strava MCP, intervals MCP. I'm actually going to add training peaks MCP because I added that myself. I think what's more helpful is perhaps I can just show you like a question I would ask this agent, right? Let's start. I'm going to check my connectors first of all. And I've got the calendar integrated. I've got training peaks integrated. ⁓ and let's also do Garmin. speaker-1: And so you tending to switch these on and off on a conversation by conversation basis. speaker-0: Yeah, because I've found that leaving them all on can bloat the context. And I actually had a problem where I had too many connections. And after like a few messages, the system would say like, I've hit my limit. need to compact the conversation. ⁓ and that's cause I had like 10 MCPs on anytime, somewhere for work somewhere for sports. So yeah, I still actively turn them on and off. ⁓ Also to save context, I would probably switch out of Opus and maybe use Sonnet. ⁓ but like, here's an example of something I can say to this agent. Yesterday, I did a yoga session. want you to analyze my heart rate during that yoga session. also want you to recap the last three or four days of Garmin metrics. ⁓ tell me how my sleep is looking, how my resting heart rate is looking. And I'd also like you to critique the run that I went on on Saturday. So this is probably going to pull information from Trading Peaks and it's going to pull information from Garmin. So here we go. I'm gathering the Garmin data and look at your yoga session heart rate. If we click in here, you can see it's looking up activities from Garmin. It's done that. It's asking permission to look at my Garmin health data. So I'll give it permission to do that. Friday's run that you mentioned. ⁓ sometimes it gets confused with dates. There we go. speaker-1: That's an interesting idea. I've been having this exact date confusion by the way this week. ⁓ speaker-0: We're doing it. It's like, it's so good at everything. And apart from knowing what time it is. Okay. So here's the yoga information. Your average heart rate was 75. You max hit a hundred. You were breathing about 15 breaths a minute. ⁓ your stress dropped massively and your body battery only dropped by six points. ⁓ this was a warm Yin yoga session. And the whole point of that type of yoga is to just relax and just allow the body to like stop striving for anything. So it looks like the data suggests that that works perfectly. It's worked. Now it's going through my sleep data, know, a good sleep score on the 30th, good sleep score on the 31st. The 2nd of February was okay. You didn't get enough sleep. And then last night, ⁓ I didn't realize that it's synced already. Okay. So I actually slept very well. ⁓ speaker-1: What a slip. speaker-0: That was like a, that was like, yeah, best sleep of my year so far. And your sleep last night was exceptional. Uh, your resting heart rate is trending nicely. It's at 47, seven day average is at 48. This is great. And your heart rate variability is in the balanced region. Uh, today's unbalanced HRV is interesting. So it's like going through everything. Uh, your body battery, you know, currently 96, you woke up at 96%. Um, and here's a critique of your run. So you did a 5.4k run, 138 average heart rate, you had a little spike, you averaged 447, good training effect for aerobic training, nice cadence. I mean, I personally don't understand how to interpret vertical oscillation, but apparently this would speaker-1: Just in case you can ask questions, right? speaker-0: Exactly. could say like, help me understand, actually, Les, help me understand the vertical oscillation. I'm not really sure what GCT means. ⁓ speaker-1: It's even telling you, is it kind of telling you off there with the critique where it's saying, hey, this was supposed to be easy, but you've gone slightly above easy. speaker-0: Yes. And so like there's a, we'll get into like the training plan underpinning this. Okay. ⁓ and we can look at how it integrates with my calendar as well. But I think I had told this agent or the training plan said like, this was supposed to be a pretty easy run and I did get. Coyed away. so, know, your form metrics were excellent. You didn't blow up and you recovered fine. So like, it's okay. Just make sure that those easy, like early kilometers aren't too fast. speaker-1: You might be running something like this on a daily basis. Cause it looks to me like this prompt in particular has basically said, look at everything I did recently. Tell me where I'm up to today. Maybe give me some insights. And then you might then use that to decide what to do for the rest of the week or how to play it. how would you use this information? speaker-0: Yeah. So that was a, guess that that demo has just shown like how much data it has access to and how quickly it can query all this. So it's like having a sort of fitness data analyst that I can speak to any time. And that was critiquing the past. A good example, here's a very good example. In the calendar this morning, I was supposed to do a Zwift ride, like an indoor bike ride. And I didn't do it because I got 10 hours of sleep and prioritized sleep. So what I can say to this, here's an explanation of vertical oscillation, by the way, ⁓ quite verbose, but if I wanted to really understand that, that would help me understand it. okay, let's, let's now help it like plan out, ⁓ a session. So I skipped my swift ride this morning and I'd like to get it done at about Could you help me plan out the best structured ride if I only have 45 minutes available? So it has access to my calendar and it has access to all my metrics. So if I was completely fatigued, it would come back and say, like, sometimes I would say you shouldn't be training. ⁓ sometimes it would say, let's do an easy ride. ⁓ but you know, it's now saying you've got a good body battery. You're not very stressed. You're in a great state for a quality training session. So it's suggesting this type of workout with this much like expected training load. Here's how I could. ⁓ I could do these main sets, do this cool down, some notes on exactly like what to aim for and like the cadence. This is incredibly accurate. I don't think my arms look big enough at the moment and I really want a gym session that focuses entirely on making my biceps look bigger. Can you create this and heavy for me? And by the way, I'm using of the things are full. speaker-1: Yeah, which is, which is super, super helpful. So I think this is interesting that it has asked you and it's, is this something that you've set up yourself to make sure that it kind of interviews you before giving suggestions? speaker-0: No, it's just quite good at clarifying. So that was quite a broad request. Let me just switch this. So there's the heavy MCP. I've just switched this on. So yeah, what equipment do you have access to? How much time do you have? What's your arm training background and any recovery considerations? ⁓ So I've got lots of arm training background. Thank you, Phil. speaker-1: Good night on Training Game. speaker-0: Okay, so I'm going to do this in the gym with all the usual equipment. I want to do this in a 45 minute session. I have been training arms a little bit ⁓ and I've just switched on the heavy MCP. So you should be able to create this for me now. speaker-1: And I think the big thing here for viewers and for people looking at how they can set up their own like programming, ⁓ is that this is not the difference to just jumping on chat, GPT free version and saying, me an arm workout program is that we're getting all of your latest data. So it's referencing your body battery. It's referencing your resting heart rate. It knows your long-term goals. So as well as trying to get hench for the beach in this situation, you, knows that you're also training for a bunch of other things. It knows. workouts you're already doing. It's pulled your calendar and said, Hey, wait a second, James, like how are you going to do this research when you've got Zwift at 4.30? And it's just super, super, super personalized. And then it's going to drop it straight into the heavy app. So you don't have to touch it. speaker-0: I mean, you've hit the nail on the head. think this demo, I don't even know if this demo really like sells how incredible this is because it has come back and said to me, I can see you've been mostly doing chest and shoulder work so far, but I don't see any bicep work. it really has. And I have been doing loads of chest work. I like, I've been doing so much of this butterfly. I love my dead hangs, but I've not really been doing much arm work. So speaker-1: called you out. speaker-0: Then it goes in and says, okay, I can see heavy has all these like, ⁓ heavy has a library of workouts and it's like, great. I can see all of these. I'm going to create a 45 minute bicep focused. How do you say this? Hyper trophy. Hyper trophy. I'm not a gym. speaker-1: That's what they say on the internet. speaker-0: All right, so here we go. It's created by set blast in my app and we're going to do chin ups, barbell, bicep curls, dumbbells, husband, hating curl, cross body hammer curl, concentration curl and cable curl. It's also suggested weights for me based on what I've done in the past. It's saying if I do seven bottom half and seven top half and seven full is that range of motion. You're more of a Jim bro than I am. ⁓ speaker-1: us. ⁓ Yeah, that means range of motion. Range of motion. speaker-0: It's actually ended my session with a pump finisher to give me an absolute pump coming out the gym. ⁓ it's told me about, you know, resting. now if I open the heavy app and I, let's just restart the heavy app, it should have sent it through. speaker-1: I think one of the cool things that you could do, so let's take this at face value, you're off to the gym to get your big arms. But what you could do is you could even get it to like teach you, you could even say like, maybe there's a couple of different approaches to this. Talk me through them first. You could get it to talk about why we're doing bottom half and top half and full range of motion and see if you agree. You can get it to... Yeah, have a look at what you've already done and factored that in. think it's a really, really, really cool way of doing things. speaker-0: Yeah, could, mean, all these questions I have, I could just go, go through this and really understand like, why am I doing these sessions? Like, well, you know, it's not just telling me to do something. It's also like educating me and like how to approach the gym. ⁓ so as is the way with live demos, this hasn't come through to the heavy app. Buster. I, I think I know how to fix that and I will show you the heavy workout after this. ⁓ and I will go and do it and I'll tell you. on WhatsApp. speaker-1: By the magic of ⁓ post-production, here is the heavy screenshot showing that this went through. And I can imagine some other use cases of this, like we're listening to podcasts all the time about training. You might hear about a new training modality that you're interested in. I might have listened to of Rhonda talk about the importance of zone two, and I might want to drop in like the PDFs around the content that they're producing, get it to go and look up the science. And then get it to factor in my situation and start to say, right, I can give one extra zone two session a week, but I'm going to get rid of one, some other session that I was doing. How might we incorporate that? Go back and forth and then, okay, go and incorporate it. speaker-0: 100%. ⁓ you could like, yeah, kick off like a deep research task to summarize something you just heard about and then have this agent just completely personalize it to your level. ⁓ you know, with, with any sport, but particularly with endurance sport, it's tempting for people to get carried away, you know, here, some influence are talking about running a hundred K a week and then just injure themselves because they've gone from like not to a hundred far too quickly. And this system would immediately push back on that and say, look, I know what you've been up to over the last like three months of your training. You're not ready to run a hundred K every week. You're not even ready to run like 50, to be honest. Let's like build this up so that you don't injure yourself. speaker-1: So one of the key, one of the things that we're seeing when we're having these conversations is like, ⁓ fact checking and making sure things are actually rooted in evidence. is as you say, listening to some guy on Instagram who's like doing 700 miles and then being like, can do that. And then personalizing, ⁓ both becoming so much more effective and so much easier to do. speaker-0: Yeah. And I want to show you something else with personalization. I did a, I did a randox blood work test in November and it was like the everyman test. Let me just, uh, let me just pull that file up for you. So in terms of personalization, I did a randox blood work test in November and it came back with this huge 65 page PDF. like the most information I think I've ever seen on my own body. And so I converted that to a markdown file that you can see here. And that is just stripping out like all of the unnecessary data. made the file size, you know, 1 % of the PDF. And now my agent, you know, can see my out of range levels, like my hemoglobin and my hematocrit. It was the first time I've said that out loud. They were both a bit low. My red blood cell count was low and that's because I'd just been on well before doing the test. My glucose fasting levels were in the pre-diabetic range, so I need to fix that. And I had this high ASO, which I think suggested I'd just had like a strep throat infection in the days before. So it explains all of this. It's got my height, my weight, you know, my everything. And now my agent has all of this. It's important to note that privacy is incredibly important. so I've said to Anthropic, do not train on my data. So I'm trusting them not to train on my urine analysis, for example. I don't think anyone really needs to know that, but let's now come back to the agent and say something like. Based on my blood work, is there anything I should be aware of when eating after this gym workout? What should I try and eat in the hours after a session like this? And it's going to give me personalized advice based on everything you can see. Here we go. Your blood work says you have a mild anemia pattern or you did in November. had some borderline metabolic markers. You just had strep infection, but everything else was good. know, testosterone, excellent humble brag. ⁓ normal. speaker-1: you can share that off. speaker-0: Yeah, this really is a sort of tech nerd slash role session. All right, so during the session, just have water. You don't need to eat anything. Then you should avoid ⁓ spiking your triglyceride by having lean sources of protein. So like a shake or some chicken breast, try to avoid fatty cuts of meat. You want some low GI carbs given your glucose issues. So try to avoid white bread or sugary drinks or simple sugars. you need to prioritize low GI carbs to avoid insulin spikes. Then, you know, this was a test that was done three months ago. I don't think I'm displaying like signs of anemia now. ⁓ but just to make sure I'm getting enough iron, it's suggesting I eat some of this. Here's an example meal that you could eat some chicken breasts, some sweet potato, big salad, some pumpkin seeds. Then, ⁓ it's getting its timings mixed up a bit, but, ⁓ before you do your Zwift session, you could have more protein. ⁓ after that, here's how you could recover with healthy fats, you know, good mix of carbs and protein. then it's sort of coming back with some principles for me to, you know, like tailor my nutrition to this blood work that I've just had done. consider some omega-3 supplements, some creatine, ⁓ you know, try to avoid these foods. I mean, This is something you'd pay hundreds of pounds a month for, speaker-1: What you're, what you're sharing us as well is just the, that blood test has become much more valuable. I think some of the other interesting things that you can do there is, ⁓ in the past I've had blood tests that I've got worried about a couple of metrics. ⁓ for example, my creatinine, I think that's how you say it has been really high and that can be a kidney problem. But, so when you, when I just get the results back, it can, it's flagged as like way out of range, but then you jump onto. Claude and it's got the context that you're kind of working out and you blah, blah, blah, blah, blah. And it's like, ⁓ no, this just means you had a heavy session within the last couple of days. Like there's nothing to worry about here. And the other thing is you can personalize those reference ranges now. So instead of just having like the reference range for everyone in the population, you can now start to say, well, actually I am this age. I have these goals. I'm doing, I have this activity level. I have this lifestyle. What should my reference range be? So for example, you're, ⁓ like, if you took not a blood test, but like VO2 max, I'm sure that your reference range that you'd be looking at wouldn't be average, right? You're not, you're interested in how do you want to be up there with the kind of endurance athletes, not with just kind of the whole of a population sample. And I think you can start to do that across lots of different metrics now. speaker-0: Yeah, exactly. And that blood work already included data on me, my height and my weight and my age. But you're right, we could go so much deeper down this personalization route. in terms of how we would then... I guess you can talk about giving agents hands, right? Like giving them tools to do things. So right now we're at the point where this agent has said, ⁓ here is exactly what you should eat. It's even saying, do you want me to design a full week's meal plan? But then I would have to put that into my fitness pile. I could add a my fitness pile integration now, or I could build my own with Claude code. And then instead of saying, Hey, do you want me to create a written PDF for you? It would go into my fitness pile. would log my breakfast, my lunch and my dinner. And then I would just open the app and like follow, follow the instructions. ⁓ so yeah, like you're getting, you know, a personalized nutritionist. do obviously have to like fact check occasionally and sort of give it up to date information, but this is, this is like so beyond what I could have done myself. speaker-1: It's amazing. It's amazing. and I think the fact checking, we, we know that with AI, do like brain first AI, second brain, third. But you do that if you went and saw a nutritionist, hopefully, or if you went and were exposed, certainly if you were just trying to collect this information for yourself from different sources, you need to engage your brain and you need to do it with AI as well. And that's okay. ⁓ thank you so much. This has been incredible. So you've shown us like how to set up a personalized fitness coach. You've shown us how to do programming with AI. You've shown us how to think about like. doing almost daily reviews and how to get used daily reviews to understand what you've been up to and what your body state is and what you should do. And now you've ended on personalized nutrition off the back of blood tests. So much to pick up on. Thank you so much. speaker-0: No problem. That was very fun. speaker-1: See you next week everyone. Don't go anywhere yet. It's Will. We have a little bonus. James sent me after we were done recording a video of the Zwift workout that got set up by the Claude process that we saw by the magic of post-production editing. Here it is and we'll see you again next time on HealthStacks. speaker-0: So I've opened Zwift and I go into workouts and then custom, and we can see this workout that Claude has just created for me. Three times eight minutes with a couple of bursts at the start, just to get warmed up. Nice warmup, easy recovery between each session. And then I cool down. And that came entirely from the AI agent that we were just talking to.