[Kamal Datta]: Welcome, everyone. I'm here with Sean Bradford today, founder and CEO of Train as One. Sean is a trained physician who eventually founded Train as One, the first anticipatory training app to help runners with their training. Sean, welcome to the podcast. I'm really excited to have you here. [Sean Bradford]: Thank you. It's a pleasure to have a chat. [Kamal Datta]: Let's start with your background. You're a trained physician, but somewhere along the way, you started an entrepreneurship journey. Can you take us back and tell us how that happened? [Sean Bradford]: People often find it strange that I've gone from medicine to technology and business. However, I actually started writing software when I was 10 years old and selling it by 13. I used that to fund myself through medical school. During that time, about 30 years ago, I was already playing with artificial intelligence, generating neural networks for disease diagnosis. I practiced medicine, working in surgical specialties, including a stint at Johns Hopkins in Baltimore. I then started using my IT skills for healthcare, eventually leaving medicine to pursue IT full-time. I helped develop a primary care GP system using technology. Over time, we expanded into different areas of healthcare, always with a focus on using AI. The transition to Train as One came when I got roped into doing a difficult multi-day mountain marathon. Trying to train for that made me want to understand how elite athletes trained. Using my scientific background, I read all the papers and found that most training was built on myth and anecdotal advice, with poor science behind it. That's when I started thinking, back in the 90s, that we could use AI to deliver something with better outcomes. [Kamal Datta]: That's fascinating. Was your transition from medicine to technology gradual, or was there a specific moment when you decided to switch? [Sean Bradford]: It was gradual. There was a crunch point where I had to decide whether to focus on the surgical side or the IT side. Even though I was in a specialty with a lot of innovation, I felt there was more rapid innovation happening in IT. I also felt that having a foot in both camps was advantageous and somewhat unique. [Kamal Datta]: You mentioned three areas: lack of scientific evidence, ineffective training plans, and injury risks from generic plans. How did you use AI to address these issues with Train as One? [Sean Bradford]: In classical training, there's a lot of guesswork and what people consider the "art" of coaching. I saw that some companies were just digitizing existing plans and calling it artificial intelligence. They might have some very simplistic rules to give the impression of being adaptive or intelligent. But this wasn't actually advancing running science at all. We wanted to solve the original problem: how should runners really train? The classic example is running injuries. The incidence of running injuries, according to studies, hasn't changed in the last 50 years, despite advances in training methods, strength and conditioning, and shoe technology. We felt passionately that something had to change, and the only way to do that was through big data. [Kamal Datta]: Can you walk us through what someone can expect if they sign up for Train as One to train for, say, the 2025 London Marathon? [Sean Bradford]: You can sign up via our mobile app or website. We ideally collect as much information as possible, including your previous running history from apps like Garmin or Strava. You'd enter your goal of running the London Marathon and provide your training settings or constraints, such as how many days a week you can run or any days you can't run. The system then generates a plan specific to you, based on your demographics, health information, running history, constraints, and goal. We can also train for multiple races with different priorities. The key point is that after each run or workout, or if you miss one, or if new health information comes in, the system analyzes everything and generates a new plan. It even considers factors like weather, adjusting your workouts accordingly. [Kamal Datta]: That's a comprehensive approach. What kind of data does Train as One use to generate these plans? [Sean Bradford]: The biggest thing we work with is your running history, as what you've done in the past is an indicator of what you can do in the future. We need velocity information during the course of runs, ideally from GPS data. We also use elevation data, heart rate, and other running dynamics like cadence. We're currently introducing running power into the model. We have a pool of data from all our users around the world, including race performances. Our machine learning model looks at these performances and the training patterns that led to them, trying to understand what patterns can lead to certain performance outcomes for individuals. [Kamal Datta]: How does Train as One handle injury prevention? [Sean Bradford]: From our preliminary work, we're estimating that we're achieving less than a 10% injury rate for individuals over a year. This is compared to the general statistic that 70-80% of runners get injured every year. Our approach involves a much different ramp-up in training volume. It's a lot slower early on, which is different from what people expect or want. We've also found that the common "10% rule" for increasing weekly mileage isn't supported by studies. The incidence of injuries is the same whether you increase by 10% or 50% week-on-week. There's a lot more nuance to it, including how you're doing those miles, at what pace, and whether you're doing speed work. [Kamal Datta]: Speaking of speed work, your system seems to recommend less of it than traditional plans. Can you elaborate on that? [Sean Bradford]: Yes, our analysis suggests that speed work is one of the leading factors for injury in training. For many people, speed work is overrated and doesn't achieve as much as they think it does. In my own experience, I recently did a 10-month marathon training block with less than 5 speed workouts, and I improved my time by nearly 10%. This goes against traditional thinking, but it's what our data and the emerging science are showing. [Kamal Datta]: That's quite interesting. Lastly, what sets Train as One apart from other AI-based training programs? [Sean Bradford]: Many adaptive planners out there use very simplistic, rules-based systems. Some are doing good things but in a very compartmentalized way, looking at specific aspects like lactate threshold or heart rate variability. We're different because we're looking at the whole person, using machine learning to consider all factors holistically rather than relying on simplistic models for individual components. We're trying to solve the fundamental problem of how runners should train, rather than just digitizing or slightly tweaking existing approaches.