fitness

· [archived] · nextjs · typescript · ai · cycling

this started as FormOS, an ai-in-the-loop endurance training platform. slowly, though, I realized that it would just better serve me as my own training system and not a larger project.

there are two overarching goals:

  1. replace the parts of TrainingPeaks I need with a dependable calendar, activity history, and season planner.
  2. build an AI endurance coach that can use that information well.

The training foundation

the app combines garmin activity data, training-load calculations, race priorities, and structured workouts. It has season blocks, plan-versus-actual history, and workout exports.

the coach has persistent chat and can propose plans or adaptations. Changes are reviewable, and completed workout history doesn’t get rewritten when the model generates a new plan.

recently i’ve been working on duplicate activity handling, accurate local dates, multiple sessions in a day, and keeping user changes intact during plan updates.

What is still missing

the platform is further along than the coach. recovery-data quality, long-ride planning, and ultra-endurance preparation still need work.

i don’t treat it as a proven replacement for a qualified coach, but the aim is to make recommendations that can be validated against my real training data and make recommendations/changes over time.