Hey — I’ve been building fitness apps with small teams for many years.
We just launched a new version of GAIN after noticing more people using ChatGPT to generate workouts. That works well in some cases, but we kept running into a few fundamental problems with that approach:
1.) Data quality & safety. A lot of fitness content online is “bro science,” and LLMs inevitably absorb that. In training, bad info isn’t just wrong—it can be unsafe.
2.) Fitness-specific UX. A gym workout is hands-free and time-sensitive. Things like audio cues, rest timing, load tracking, pacing, and exercise sequencing depend on context and capability ChatGPT doesn’t really have (e.g. squat vs curl timing, user goal, fitness level, exercise preferences, injuries, fatigue, etc.).
3.) Missing location/equipment context. Over time, we want to know what equipment is actually in a given gym or apartment setup, incorporate geo-location, and generate around that. Vs. asking users to re-prompt every time.
Tying together a bunch of expert fitness knowledge, we built a rules-based + data–driven workout generator for personalized, adaptive training. It pushes one clear, daily workout and adjusts when you miss days, travel, or change constraints—without requiring you to “start over” each time.
This is early v1. We haven’t turned on many feedback loops yet (tracking, soreness, preferences, etc.), but those are coming. We’ve already mapped out generation for cardio, yoga, mobility, and recovery, including how to balance them over time and according to various fitness goals, user prefs and usage.
I hear people trying to get CGPT to do weight tracking and progressions, but there's so much bogus data to sort through and not much published, so we think we'll do that much better, too, collecting real-world data with a purpose-built tracking UX.
Would love feedback on the approach and the app itself. App is live on iOS and free for now.
nickg•3w ago
We just launched a new version of GAIN after noticing more people using ChatGPT to generate workouts. That works well in some cases, but we kept running into a few fundamental problems with that approach:
1.) Data quality & safety. A lot of fitness content online is “bro science,” and LLMs inevitably absorb that. In training, bad info isn’t just wrong—it can be unsafe.
2.) Fitness-specific UX. A gym workout is hands-free and time-sensitive. Things like audio cues, rest timing, load tracking, pacing, and exercise sequencing depend on context and capability ChatGPT doesn’t really have (e.g. squat vs curl timing, user goal, fitness level, exercise preferences, injuries, fatigue, etc.).
3.) Missing location/equipment context. Over time, we want to know what equipment is actually in a given gym or apartment setup, incorporate geo-location, and generate around that. Vs. asking users to re-prompt every time.
Tying together a bunch of expert fitness knowledge, we built a rules-based + data–driven workout generator for personalized, adaptive training. It pushes one clear, daily workout and adjusts when you miss days, travel, or change constraints—without requiring you to “start over” each time.
This is early v1. We haven’t turned on many feedback loops yet (tracking, soreness, preferences, etc.), but those are coming. We’ve already mapped out generation for cardio, yoga, mobility, and recovery, including how to balance them over time and according to various fitness goals, user prefs and usage.
I hear people trying to get CGPT to do weight tracking and progressions, but there's so much bogus data to sort through and not much published, so we think we'll do that much better, too, collecting real-world data with a purpose-built tracking UX.
Would love feedback on the approach and the app itself. App is live on iOS and free for now.