Title: VeloForge — adaptive coaching that connects training, recovery, and fueling
Hi everyone,
I’ve been following the projects in this category and wanted to share what I’ve been building.
I’m Kerry—a former professional triathlete, longtime road cyclist, and still training seriously. Over the past year, I’ve been working on VeloForge, an adaptive AI cycling coach for athletes who want structured progression but also deal with group rides, fatigue, work, travel, and missed sessions that constantly break static plans.
Several AI coaching tools are appearing now, so I want to be clear about the difference I’m trying to create.
VeloForge is not an LLM that simply reads activity data and invents workouts. The AI sits on top of a deterministic coaching system with defined training rules, physiological guardrails, and structured calculations.
The core system includes:
-
Adaptive training phases based on athlete level, goals, and race timing
-
Individualized power targets derived from the athlete’s power profile
-
Weekly limits on productive hard volume
-
Daily adjustments using sleep, HRV, resting heart rate, readiness, and completed training
-
Automatic schedule rebuilding when workouts are missed, moved, or replaced
-
Recognition of group rides and their actual intensity cost
-
Race-specific build and taper logic
-
Daily calorie, carbohydrate, protein, fat, and ride-fueling targets
-
Cycling-specific strength work aligned with the training phase
One of the main problems I wanted to solve is what happens when the original plan is no longer appropriate.
For example, an athlete completes a hard five-hour group ride on Saturday, accumulates significant high-intensity work, sleeps poorly, and still has a threshold session planned for Tuesday.
A static plan continues as written. A generic AI coach may improvise.
VeloForge should account for the hard work already completed, the athlete’s remaining weekly intensity budget, recovery status, training phase, and upcoming sessions—then adjust the week without destroying the purpose of the block.
Training and nutrition also remain connected. When a workout changes, VeloForge automatically recalculates daily energy needs, macros, ride fueling, recovery nutrition, and the remaining meals for the day.
Current status
VeloForge is a Flutter-based iOS app distributed through TestFlight.
-
Apple Health and Wahoo are supported
-
Strava integration is live
-
Garmin access is pending approval
I’m still actively debugging and refining the system, so this is not a finished product.
I’m looking for a small group of serious cyclists to test it and challenge the coaching logic.
The ideal early tester:
-
Trains approximately 8–15+ hours per week
-
Uses structured workouts but also participates in group rides
-
Tracks power and heart rate
-
Cares about recovery and fueling
-
Has experienced plans failing to adapt when real life changes
-
Is willing to provide direct and detailed feedback
I’m particularly interested in feedback on:
-
Whether the weekly structure makes physiological sense
-
How the system responds to missed or changed workouts
-
Whether recovery adjustments are appropriately conservative
-
Whether fueling recommendations match real training demands
-
Where AI explanations are useful versus unnecessary
-
Any situation where the system makes a clearly poor coaching decision
I’m building VeloForge because I believe AI coaching is only useful when strong reasoning is combined with clear training rules and reliable calculations.
Anyone interested in testing it can join here:
I’m also happy to discuss the architecture and coaching approach in this thread.
Honest criticism is welcome—this community understands the difference between a polished demonstration and a system athletes can actually trust.
