Workout Lens v1.2 is live. It connects to Apple Health and helps you understand workouts, recovery, training load, and readiness. I also added AI Coach Pro: private on-device coaching where supported. Looking for feedback from Apple Watch runners.
@Google@googlegemma Already running Gemma 4 e2b on-device in an Apple Watch runner app. The win isn't benchmarks โ competitive reasoning on a phone with zero network round trip is the actual unlock. Training data never leaves the device. Private health AI just got viable.
@Gathiyo 100% this. The Apple VO2 model is trained on runners running, then handed to a cyclist who looks like a sloth on paper. As a runner I love the bias, but for cross-sport athletes it is genuinely misleading. Whoop handles it better by using resting HR trends, not workout pace.
@therunninggenie Congrats on 400. Apple Health is the easier integration - Garmin Connect is where the pain lives. How do you handle the merge: same source or let user pick canonical? I went the other way - Apple HealthKit only, and that decision is rough on Garmin loyalists.
@jordantcarlisle HealthKit is great until the export crashes on multi-year datasets - same problem I ran into. The deeper issue: every app reads the same HealthKit data but writes its own format. The API exists, the open standard does not. Apple will not open it. Builders do it themselves.
@sarahMo3W Classic Apple Watch rounding. The 5K celebration triggers way too early. Curious if you train by distance or time - I have found runners care more about pace than the cheer milestones. Either way, that should have been a 10K post.
@Thee_KaylaT Three-device stack is real for serious runners. One thing I keep coming back to: HealthKit is the only hub that pulls Apple Watch + Garmin + Whoop together, but no app surfaces "you are cooked, rest today" cleanly. The recovery signal is there, the UX is not.
@dksf Curious - how are you handling the recovery side? I am building on the same idea (Apple HealthKit on Watch) and the hardest part is telling runners "today is a rest day" in a way they actually believe. Daily readiness or weekly load?
@alexyuminjung Interesting approach combining HRV + sleep for readiness scoring. One thing I keep coming back to: weekly load trends vs daily readiness - runners tend to optimize the day, then blow up by Sunday. How are you handling the weekly view?
@TraffAlex On-device is exactly where health/fitness should go. HR, recovery, RPE โ none of it needs to leave the device. Building around that for Apple Watch runners: coaching that runs locally where supported, so your training history isn't quietly living on someone else's server.
@NixFred This is great. The "illness 48h early" insight is exactly what iPhone Health never surfaces. Same aggregation-logic pain on my end. I went the focused iOS route: turn Apple Health workouts + recovery into one clear "train hard, easy, or recover" decision.
@Kevin_McE_ Great breakdown. Most apps use the Apple Watch calorie number as a flat input โ they miss the actual session structure (intensity, duration, HR zones). Pairing RPE + total volume + HealthKit bodyweight trend with daily energy burn is the cleanest adaptive engine I've seen.
@superageapp Readiness as recovery debt is the right framing. Most apps show HRV in isolation instead of layering it with the actual training load that caused it. Single HRV reading = noise. The 7-14 day trend vs your chronic load is the real signal.
Looking for 10 Apple Watch runners to test Workout Lens.
It reads Apple Health workouts and turns them into recovery-based training advice.
Want to know:
- is the advice useful?
- is the paywall clear?
- is export valuable?
- what's missing?
Reply and I'll send the link.
Workout Lens v1.2 is live. It connects to Apple Health and helps you understand workouts, recovery, training load, and readiness. I also added AI Coach Pro: private on-device coaching where supported. Looking for feedback from Apple Watch runners.
5 sprints, 26 tasks, zero shortcuts.
DealOS: agent-native C2C marketplace where AI agents search, negotiate and close deals for you.
LLM intent extraction, trust scores, MCP server, escrow โ all shipped.
Building the anti-Marktplaats. ๐๏ธ
#buildinpublic#AI#marketplace
5 sprints done on DealOS. All features shipped. Then we tested the full flow. Individual features work does not equal the whole journey works. The gap is integration. Next: end-to-end hardening until real users close deals. #buildinpublic
36 features shipped. AI matching, trust scores, iDEAL โ all done.
Then I tested the live site. Core search flow breaks.
Green checkmarks don't ship working products. This is the real build.
Back to fixing. ๐ ๏ธ
#buildinpublic#marketplace