Featured in TalTech's Trialoog: how KULG combines science and AI on the running track
TalTech's science publication Trialoog just published a deep dive into KULG's AI and the story of how it came together.
Two people solving the same problem
Leading KULG's product and AI direction is Martin Tamm, a PhD researcher at TalTech's School of Information Technologies Department of Software Science and a runner with 13 years of training experience. Alongside his research on digital product passports in e-commerce, Martin had built his own AI-based training tool - one designed to create a science-based, adaptive plan around his training history, load, recovery, and health data.
The tool sat unused for a while; turning it into something more would have taken time and resources he didn't have at the time. Then, in February 2026, an introduction connected him with the KULG team and it turned out we were solving the exact same problem: making training plans more personal and more grounded in science. Since joining, Martin has led the growth of KULG's AI and its knowledge base.
Research no coach could read alone
No coach has time to keep up with thousands of research papers on training, recovery, and injury prevention. Especially as new studies keep refining, and sometimes overturning, long-held assumptions.
“KULG’s knowledge-base agent has analyzed over 2,200 running and training-related scientific studies in the past six months.”
KULG's knowledge-base agent has analyzed over 2,200 running and training-related scientific studies in the past six months, filtering for high-quality, evidence-based research and keeping the knowledge base current as new studies emerge. As Martin put it: "No ordinary person or coach could get through that volume of research articles."
Why not just ask ChatGPT?
A generic AI assistant can produce a training plan too, but it doesn't know your last few months of training, your health, your recovery, or your injuries. It's also not guaranteed to hold onto everything you've told it earlier in a long conversation.
KULG works differently: it keeps every athlete's training, health, and recovery data in a persistent profile, so recommendations are built on real history rather than a single chat. And when the AI doesn't have enough information, or a recommendation lacks a scientific basis, it's designed to ask rather than guess.
Every runner is different
KULG doesn't treat every runner the same, because the same pace can mean very different things for different people. Martin often points to a race where he ran alongside one of Estonia's top runners: at a similar pace, the other runner's heart rate sat around 150 bpm, while Martin's was 186.
That's why KULG builds a picture of each athlete individually - training history, goals, resting and max heart rate, body composition, sleep, and recovery data, rather than applying one plan to everyone.
Read the full post
The article goes deeper into how KULG's AI knowledge base works, and what the team is building next.
