How I Built an AI Fitness Company From a Hospital Graveyard Shift

4:17 AM. Jake, 40, hospital security supervisor, sitting in his Ford F-150 in the staff lot. 308 pounds. Fourteen hours on his feet. Ferritin of 7.

Drive-thru breakfast on the passenger seat because the cafeteria locked at midnight. The third fitness app of the year just got deleted off the phone in his hand. That app assumed 7 AM workouts. Sunday meal prep. Eight hours of sleep.

It assumed a person who did not exist.

TL;DR

• I went 308 → 196 on hospital graveyards. No $300/mo trainer. No app that fit.
• Shift workers carry a 25–40% elevated metabolic syndrome risk over day workers (occupational health literature). The standard program is the wrong tool.
• Legacy In Motion is the system I built for myself, productized: Chiron (AI head coach), HERMES (scrapes ~12,000 papers/week), and a daily program-update worker that rewrites your week off your HealthKit logs.
• Per-meal protein has a leucine threshold (Moore 2009 AJCN; Witard 2014 AJCN; Murphy 2018 BJN). HRV-guided training beats fixed periodization in disrupted-sleep adults (2025 meta-analysis).
• Static markdown blog. One product. One link. If you never pay me a dollar, the science still works.

The average person does not exist

The conventional fitness industry serves a hypothetical human. Eight hours of sleep. Predictable schedule. Sunday meal prep. A relaxed 24-hour window before the next squat day.

That person is a marketing fiction. The actual people paying for programs are moms on three broken hours, ER nurses cycling rotations, contractors home wrecked at 7 PM, and hospital security supervisors eating from a vending machine at 4 AM.

The shift-work tax is not a mood. Occupational health literature puts metabolic syndrome risk 25–40% higher in shift workers versus day workers. Same body. Different clock.

Generic programs do not fail because the science is wrong. They fail because the science was modeled on a person who is not you.

When I missed a session at trainer number four because I had been awake 22 hours, the trainer marked me "non-compliant." That is not coaching. That is a gradebook with a logo.

The protein math nobody told me

For a year I ate "high protein" and lost nothing but money. The math was off and the apps did not flag it.

Per-meal protein has a leucine threshold to trigger muscle protein synthesis (Moore 2009 AJCN; Witard 2014 AJCN; Murphy 2018 BJN). At 308 pounds with a wrecked sleep window, my hits were too small and too spread out. The leucine load never crossed the line.

The fix was not more protein. It was concentrated protein at the right beats — including the meal that lands at the end of an overnight, where most programs tell you to "just sleep."

You can be in a calorie deficit, eat 180g of protein, and still not signal muscle to stay. The signal is per-meal, not per-day.

This is the pattern Chiron — our AI head coach — flags in the daily program review. Log a 2 AM meal, watch the sleep score crash the next morning, and Chiron rewrites the lift block before you wake up.

HRV beats the calendar

The next thing the apps got wrong: they planned my week on Sunday and then refused to redraw it.

A 2025 meta-analysis on HRV-guided training in adults with disrupted sleep schedules ran the math. When recovery markers drive the load, you adapt faster and you blow up less. Fixed periodization assumes a recovery curve shift workers do not have.

When your Apple Watch logs sub-90-cadence walking all week, the daily AI program-update worker writes a three-block cadence prescription before it touches your resistance calendar. The watch told the system you were under-recovered. The system answered before you opened the app.

Walking is good. Walking with a system listening underneath it is the difference between maintenance and remodeling.

The stack is boring on purpose

That is a feature, not a bug.

Coaching layer. Chiron handles onboarding, program generation, weekly check-ins, and protocol changes. Trained on my actual 308-to-196 drop, not a stock prompt. Knows what 22 hours awake does to recovery. Does not shame a missed session.

Research layer. HERMES scrapes roughly 12,000 fitness and physiology papers a week so your protocol updates the moment new evidence lands. You do not wait for a trainer to read a study. The system already did.

Automation layer. Check-in reminders fire on your schedule, not a fixed 9 AM cron. The voice-note check-in catches the cortisol-tell in your voice before the scale moves. In-app meal log plus barcode scan handles the "I do not have time to track" problem in one tap.

App and site. Static-first. The blog you are reading is markdown in a git repo. If a content update takes more than two minutes to deploy, the system is too complicated.

The whole stack runs on a fraction of what a traditional fitness company burns on software. The savings go into keeping the program accessible, not into a runway.

The four decisions

Specific, not broad. I could have built a generic fitness app. Instead I built for the customer the spreadsheet pretends does not exist. Chaos schedules. Thyroid crashes. CIDP. PCOS. Three failed normal programs in the rearview. The specificity is the moat.

The story is the product. No faceless startup with a chatbot and a stock photo. Every protocol is grounded in a real cut from a real human who slept four hours a night and dropped 112 pounds anyway. That cannot be cloned.

Do not overengineer. Tech for its own sake is a trap. Every feature has to serve the person on the other end. The voice-note check-in ships. A social feed does not.

Content as infrastructure. This blog is not a funnel. It is a public record of what we actually know. If you follow every post and never pay me a dollar, the science still works on your body. That is the bar.

Who this is actually for

The 41-year-old contractor who gets home wrecked with 20 minutes before three kids land on him.

The 36-year-old mom whose thyroid meds change every six months and whose last trainer told her to "just be consistent."

The accountant who has tried four programs and still weighs more than he did a decade ago.

112 pounds is not grit. It is a system. I am handing you the system.

The fitness industry has always served the people who already know how to win. I am building for everyone else. If that is you, start here.

— Jake Founder, Legacy In Motion