Latest / Healthier: Building Health & Fitness Businesses / The best vibe-coded health system we've ever seen | Mike Hudack, Sling Money | Healthier Ep 4
Transcript
- speaker-0: I have a full-time job running a startup and I have three children and I've like kind of built this thing just on the side with a prompt here and there and that is wild. That's probably one of the most exciting things to happen in the last hundred years. Like why build a UI for yourself to then... make inferences, you know, analyze that data when you can just have a bot analyze all the data for you. It'll do a better job. I started asking questions about my last year and it told me that I was overtraining in September and it literally told me it was like you can either lift or cycle to and from work on a given day. Don't do both. And it pointed to like my OroRing data and HRV, but also to blood tests and everything else. And it was like, listen, you know, this two week period or whatever, you just overdid it. can't do that, you know? And it was like very specific and very prescriptive and I think very correct. speaker-1: Welcome back to HealthStacks. Very, very excited to be joined today by product legend, Mike Udak. Mike has been behind products that almost everyone watching this would have used, be it at Facebook, Deliveroo, Monzo, and now his own company, Sling Money. But we are not going to spend a huge amount of time talking about those things because Mike has been tinkering around with all of the tools that are available to all of us. ⁓ to use them in some really, really cool ways to help them fitness. So Mike, welcome to HealthStacks. And can you tell us, tell people what type of things you've been playing around with and what they might expect to get out of today? speaker-0: Well, I basically, maybe a few weeks ago, started playing around with Claude Code. the first thing that I built was like an analyzer for my DNA. I went and my whole genome sequenced. I just, there are no good tools out there for understanding your genome. And so I built a thing with Claude Code to do that. And then I realized that that gave me an incomplete picture. And I wanted to integrate some blood tests. And so I did that. And then I thought, ⁓ I bet my aura ring has an API. And so I integrated that. And then I started importing all of my workout data and I'd been weighing myself on a wide thing scale for like 11 years. And I integrated that. And then I built a food, like a food tracker, like a nutrition tracker. And then I built my own workout tracker. ⁓ And then a dire, then like a little diary thing. And then I wrapped it all in an MCP server. And the thing I'm trying to do today actually is get OpenClaw talking to it properly and be able to converse with it over WhatsApp, which isn't totally working yet. It's kind of like a little bit up and down. speaker-1: Yeah, you've got to that point that we've got lot of us have got to, which is like just one, just one more prompt bro with Claude code and just, just if I wonder if, then usually it can. speaker-0: I mean, it's absolutely ridiculous. I built the food tracker at about 10 minutes and then I had it wired up to this database provided by these wonderful guys. The company's called Fat Secret and they operate a really nice nutritional database. Quad found it, but they lock it to the U.S. and I live in the UK and I emailed with them and asked them for access to UK. They wanted 1500 bucks a month for that. So then I thought... ⁓ I think I know how to solve this. And so I just gave it my open AI key ⁓ and just built a thing myself where you can describe in free text what you're eating in it. QC, the macros and everything. speaker-1: What's, what would you say has been like the number one most useful thing from this amazing kind of all singing, singing or dancing health assistant that you've now built. When you find yourself using it day to day, what's, what's impressing you? What are you going back to? What's been most useful? speaker-0: The big thing that was kind of amazing was I started asking questions about my last year and it told me that I was overtraining in September and it literally told me it was like you can either lift or cycle to and from work on a given day. Don't do both. And it pointed to like my OroRing data and HRV but also to blood tests and everything else. And it was like, listen, you know, this two week period or whatever, you just overdid it. can't do that, you know? And it was like very specific and very prescriptive and I think very correct. ⁓ It's given me a lot of stuff like that and then it's also helped me really understand like body composition changes over time based on what you eat and like all that kind of stuff. With a lot of these things, I'm not sure for how long you keep up the habit of doing all these things, but you you gain intuition. ⁓ And then with the DNA, I've learned a ton about myself and... risks for different things. like, you know, my ⁓ bilirubin levels in my blood are, you know, consistently elevated. And it turns out that's because I have the thing called Gilbert's, which is basically harmless, but you know, you can actually see it in my genome. speaker-1: feel like the value of these things is really interesting thing that's happening is that the value of a blood test to us has shot up because we can then go like to the average consumer has shot up because we can go and build on top of it. I wonder, but I, but I wonder separately how much of that value actually accrues to the people selling us the blood test because usually we just kind of take it and off we go. Like I've tried a bunch of these new sexy ways to get blood tests and they've all got sexy dashboards. And the first thing I do is kind of export CSV, take it off to my own thing. It's an interesting new world. speaker-0: Yeah, for sure. mean, I think that like all of those kind of inputs are probably going to become commoditized over time is my guess. ⁓ Very briefly, I believed that like nutritional databases would become more valuable in this world. And then I realized I could just like build my own in literally 10 minutes. And so I no longer think that. ⁓ But you can't analyze your own flood. So you do have to pay somebody to do that. do need to use a lab. It's just like, they're not going to accrue more than $50 worth of value. speaker-1: And just to talk us through the steps you essentially started off Claude code. What happened next? speaker-0: Yeah, I mean, I had a very large VCF file from ⁓ DNA complete.com, which took about nine months to get, which is a story in and of itself. And I couldn't figure out how to analyze it. And I just ended up asking Claude, you know, hey, help me understand my DNA. And it offered to cross reference it with SMPedia. And I was like, great, do that. And then when it finished doing that. It was like, do you want me to cross-reference it with any other databases? And I was like, well, like what ones? And it was like, oh, well, the NIH operates these two databases of clinically significant findings. And there's a UK database of drug response based on your genome. Do you want me to link those? And I was like, sure. And before we knew it, we had like six cross-referenced. And then I kind of had a UI to navigate all of that. But then I decided that that actually didn't really make sense. And the best interface for this would just be like talking to a bot. And so that's when I rolled the. speaker-1: That's really interesting because a lot of the people I'm speaking to, the UI is one of the attractive things, right? It's like, don't just have to talk. You can build up some pretty thing that's kind of you dig in, you click in, but it's interesting you've taken the approach to. So you went UI and then back to to a bot. speaker-0: I, yeah, I use both that I have a UI that works on desktop and mobile. And I have a, you know, like a web app on my phone that you just open up and it works great. but I find that for analyzing, for finding relationships between, you know, blood tests and DNA and, you know, all this kind of stuff, why do the work yourself when you can have a bot do it? Like why build a UI for yourself to then make inferences and analyze that data when you can just have a bot analyze all the data for you, it'll do a better job. speaker-1: And is there an example of how you get it to the point where it's doing a better job or like some principles that you should, that people should bear in mind if they're trying to go away and set this type of thing up themselves. speaker-0: Well, I think the danger, I think the one danger, so if you make all of this data available to a bot, and I found that there's a really big difference between Chatch GPT and Claude here too, is you can easily drive yourself crazy. Like I, you know, I started going down this road of like asking about my ⁓ dementia risk. It's one of the reasons why I did the genome sequencing in the first place, because I just wanted to understand that. And... You know, it came back and it was like, ⁓ you're at higher risk for like Lewy body dementia. And I was like, great. What should I do to, you know, avoid that or whatever? It was like, ⁓ you know, it's basically the same list as for everything else. Like, you know, exercise and eat well and sleep well. And then it was like, ⁓ I noticed that your sleep quality has degraded the last two weeks and your restlessness score is poor. Have you considered whether or not you might have like this? You know, REM sleep disorder, which is an early warning of movie body dementia. And I was like, that's pretty fucked up, dude. speaker-1: You need to start connecting into your conversations to see if there's anything that it's given you that would cause the restless sleep. speaker-0: For sure. And I think you just have to be very careful about that sort of thing. And I've noticed for whatever it's worth that like ChatGPT does that a lot less than Claude does. Claude is like very jumpy. It's like a first year medical resident, whereas like ChatGPT has been practicing medicine for 40 years and is like, don't worry about this. It's fine. So this is the desktop version. mean, it scales down to mobile. can imagine what it's like, but you've got You know, all of the aura data here, ⁓ you've got, you know, body fat percentage and weight, ⁓ food log, ⁓ diary habits, you know, workouts and blood tests on the dashboard. And you can do stuff like, you know, I don't know. speaker-1: So you've replaced Cal AI or my fitness power scanner instantly with this. speaker-0: Yeah, and so you can just like log whatever you want. You can also like take a photo or scan a barcode. Yeah. You know. speaker-1: this is where it's going to start flashing and I really hear the sirens go off. And just to reiterate, that's now looking up, that's not going to a $1,500 a month. So that's going off to what you've set up with. speaker-0: No, no, that's Yeah, it's literally just like the OpenAI API. ⁓ And so then, I don't know, we can get rid of this. You've got all of the history of your Aura data here. ⁓ And you can play around with this and look a year back, or I don't know how far back all time goes. speaker-1: Were you able to pull all of this historic data through, the MCP or through the API? Yeah. So, so once you had the UI in place, it could pick up everything historically as well, which I think is so powerful, right? Because for people that have been wearing a ring or a strap or whatever been like on the withing scale for 14 years, suddenly you've got a new way to interpret all of that data. speaker-0: Here you go. mean, this is going back 13 years, right? And you can. speaker-1: You're too amazing. Guys, this is great. There's a product. There's definitely a product. speaker-0: I think there's a thing there, And so then you can go into DNA. ⁓ So this is, I didn't spend a ton of time on this UI, like I said, because it's got chat, ⁓ you've got everything from S and Pedia, clinically important ⁓ variations, ⁓ traits, drug impact. speaker-1: Did you come into this, how much did you know about DNA and how to interpret DNA before you built this? Are you getting it to kind of teach you as well or are you just reading the raw data? speaker-0: Yeah, I guess like a little bit of both. ⁓ I mean, I'm like vaguely familiar with the way that this stuff works, but ⁓ I knew about like SMPedia and all that kind of stuff. So I knew enough actually to do that. ⁓ Claude code over time prompted me to like, you know, integrate additional databases and everything. And then you learn as you talk to it about. You know, you learn to ask questions like, okay, I have this gene that influences this risk. And I suppose, you know, you can either approach that. I think that your goal should be to understand how a genetic counselor would talk to you about that. So you start asking questions like, ⁓ okay, I have a 10 times increased risk for this. What does that actually mean? What are my actual odds of developing this? You know, okay, fine. Based on everything else, you know, about my lifestyle and, and, you know, general level of fitness and everything like. now adjust that again. And one of the things that I ended up building actually is, you know, I probably won't show you this, but of course I ended up building a kind of like a family history and all that sort of thing. So you have, you know, I was just like, okay, let's put everything in here that a doctor would. speaker-1: And are you then, the super interesting as well, cause you can then start to ask it like what would be helpful to help you see the full picture and it can say, to help you be as effective as possible in the same way that we do that with like, when we give Claude a task in general, it's like interview me first. You're almost able to do that, but then put it into the persistent history and say, and here's my family history. And then someone, somebody else can reference that later on. speaker-0: 100%. ⁓ And so I've just kind of like added all of those things over time. And then, you know, I was like, ⁓ I've been recording workouts like in, you know, this app called strong for a long time. ⁓ cool. Like let's bring that in and then you can see kind of like all this coming together. And this is what I mean by like, at some point you lose, you could spend a year developing the, you know, the design to help you yourself. find all of the relationships between these things and understand how they work together. Or you can just ask the bot to do it. I think it's just like way more effective to ask the bot. speaker-1: Have you managed to get it to the point where it's kind of proactive, where it's coming to you with interesting things? speaker-0: Well, I think that's an appeal of open Claw, where the way that that works is you can talk to it on WhatsApp, and you can give it tasks, and it can run Cron jobs and all that kind of stuff. so my plan is to ask it. We do this at work now. So we have a bot on our Slack that has access to our Metabase and our BigQuery and our internal admin tooling and all that kind of stuff. posts a summary every day of like, here's what happened yesterday. Here's what growth looks like. Here's what engagement looks like. Here are the things that you should be worried about and here are the things you should be excited about. And I think it would be kind of cool to build something like that personally. And I want to build a health version of that in OpenCla. But I also want to build, I'm trying to figure out how to get it like my favorite restaurants and, you know, hook up like 11 labs to give it a voice and. allow it to make phone calls and be able to say, hey, know, book me something, book me dinner tonight, you know, or whatever, as well as having it ping me in the morning and be like, hey, you should take a rest day today. speaker-1: That's amazing. Are there particular tools, be they physical, be they apps that you've found that played really nicely with this setup? has it led you to go out to try and get more data? Like, do you find yourself thinking, if I just go and get this extra test, or maybe I start wearing this extra thing, then it's going to get even more useful. speaker-0: So Aura is amazing. The Aura API is phenomenal. ⁓ It's great, and they make everything available. I can't recommend it enough. I got a whoop band and tried it, and I'm returning it because they make a tiny subset of their data available through the API. ⁓ it just, I don't know, I don't think it adds anything. ⁓ Why things is great, like the scale data is incredible. You saw, you know, like 14, 15 years of data there or whatever. I also think that they're like blood pressure monitor is really good. So I didn't show you, I have blood pressure, blood pressure monitoring in there as well. ⁓ So those are great products. I think they cover most of what you need. Like the ring is giving you HRV, resting heart rate, you're getting workout logs, ⁓ you're getting weight and body composition, which moves around a ton, so you have to like do a bunch of work to kind of smooth that out. ⁓ I don't know that you need that much more. mean, you know, go and get a blood test and upload the PDF. speaker-1: That's the first place. That's the first place for everyone to start. Right. If people haven't done anything, that's the way to start is to take the kind of tests that you get that gives you just your number and a reference range that's for the whole population. And you tell it, here's who I actually am. I'm not the whole population. Here's my, where I want to be. And like, give me some analysis and some reference ranges based on like the type of person I want to be. Not, not the kind of average. that's, that's kind of first step. ⁓ speaker-0: Yeah, yeah, I think that makes sense for sure. speaker-1: And just finally, a couple of questions. were you ever using the Aura AI coach advisor beforehand? And has this changed how ⁓ likely? speaker-0: I have never used it. I mean, I've tried to use it. I just find it very dissatisfying. don't find it to be that smart. ⁓ So this is doing for me what I had hoped it would, what the Aura product would do really. Yeah. speaker-1: That's super interesting. think that echoes the experience of a lot of people and people are essentially going out and building this stuff. And again, it makes it difficult for those product teams at these larger companies to be able to build the type of stuff that you're willing. And then just finally, a little bit of navel gazing. Is there anything that you're particularly excited about in the world of health and fitness and performance technologies? We look into the next, into the near future. speaker-0: I don't know, man. think that like the most exciting thing that has happened, there are two really, really, really exciting things that are happening in the world right now. One of them is AI and the fact that, you know, I have a full-time job running a startup and I have three children and I've like kind of built this thing just on the side with a prompter in there and that is wild. And that's probably one of the most exciting things to happen in the last hundred years. And, and then the other thing that I think is incredibly exciting is the way that money is changing with blockchains and stable coins and tokenized assets and everything. And, ⁓ those two things are going to transform the world in really fundamental ways. ⁓ I think that it turns everything else like in health into kind of a signal generator, you know, and it's a hard problem. or it could decide to shut down the API in order to deal with this problem, or they could decide that actually they have, and I think this is true, the best hardware for collecting all these biometrics. And I'm totally happy, thrilled in fact, to be paying 10 bucks a month for that or whatever. And I don't think you need that much. I think you need a company that does that really well, and you need a company that does scales really well, and then you probably... I feel like I'm missing blood glucose monitoring. I'm excited about advances there. And I have some friends who are working on like a patch type thing, like little transdermal sensor for a bunch of additional markers, you know, not just glucose, but like a bunch of hormonal markers and all that kind of stuff. I think that's going to be very exciting. speaker-1: Yeah. Amazing. Mike, thank you so much for showing us what you built. It's incredible. I think it's going to get a lot of people excited. A lot of people spending late evenings with Claw Co. trying to replicate it. Maybe a lot of people reaching out to you trying to register as well. ⁓ It's been a pleasure. Thank you so much. speaker-0: Thank you, Will. See you later.