Hi HN, I’m a self-coached endurance athlete and long-time user of Intervals.icu. Over the years I’ve gotten comfortable interpreting my own training data (CTL/ATL trends, HRV, fatigue, etc.), but I kept running into a practical problem: The training plan makes sense when you write it, but real life rarely cooperates. Bad sleep, work travel, missed workouts, or suddenly having only 45 minutes instead of two hours. In those moments the question becomes: does the planned workout still make sense today? I built a small tool called PacePartner to help with that: The idea is not to replace coaching or generate perfect plans, but to act as a decision layer on top of Intervals.icu. It connects to your account and reads things like: training load (ATL / CTL) HRV / recovery signals sleep data planned workouts upcoming races You can then ask questions like: “Should I still do threshold today?” “I only have 60 minutes — what should I train?” “I missed yesterday’s workout — how should the week adapt?” If the workout changes, it can also push the new session back into the Intervals calendar. What surprised me while building this Initially I assumed the main challenge would be building a very specialized “AI coach”. In practice the biggest improvement came from simply giving the model good context from the athlete’s actual training data. Most athletes already have a training plan. The useful part isn’t generating one from scratch — it’s helping adjust it when circumstances change. Rough architecture Intervals.icu OAuth integration Pull training metrics + calendar data via API Contextual prompt layer grounded in common endurance training principles Conversational interface (web + messaging) Still early and very much a work in progress. Would especially appreciate feedback from: endurance athletes who use Intervals / TrainingPeaks people building AI assistants around structured datasets anyone thinking about AI systems that augment decision making rather than automate it Happy to answer any questions. |