A Weekly Health Review with an LLM

People ask how to optimize health with AI as if the answer is a supplement stack. The boring version works better. Sleep enough to train. Keep most endurance work easy. Eat for the sessions you actually did. Watch overnight HRV against your own baseline. Then change next week's intensity when the trend says so.

Walsh and colleagues (2021) put sleep in the middle of athletic performance, not at the edge of it. Fullagar's earlier review is the same story from the other direction: sleep loss hits exercise and the responses around it (Fullagar, 2015). If your Sunday chat does not start with sleep duration and a rough bedtime, it is already looking at the wrong lever.

Sleep: duration, and whether bedtime drifted. Not four-stage pie charts. Overnight HRV and resting heart rate versus a four-week baseline. Plews treated HRV as a window on adaptation over time (Plews, 2013). Kiviniemi showed you can let that window change whether the day's endurance work is hard or easy (Kiviniemi, 2007).

Training: what you did, RPE, easy versus hard. Seiler's distribution is still the map I want the model to protect: most of the running stays easy (Seiler, 2010). Food: protein and enough energy for the work, in the sense of the ACSM joint position on nutrition and athletic performance, not a grocery list of powders (Thomas, 2016).

Mood and life: travel, alcohol, a sick house, a deadline. The wearable will call that strain or a bad night. The note tells the model why. A 2026 biosensing review is honest about the role here. Algorithms can support coaching. They do not know the deadline (Biosensors, 2026).

One primary quality to protect next week. Strength, or a quality run, not both dressed as a peak. One maintenance quality so the other quality doesn't vanish. The days that stay easy, named. If sleep was short three nights, the first lever is bedtime, not a new interval set. If HRV is down and strain climbed on the run days, cut the hard endurance before you cut the lift you still need.

A bad reply diagnoses you from Wednesday. A good reply talks in sevens. It says your seven-day sleep average slipped, HRV followed, and Tuesday's intervals are the session to move. Then it stops. You do not need a 40-point wellness program. You need a week you can finish.

Do not re-query the model every time recovery turns yellow. That teaches you to treat a wearable like a mood ring. Sunday is the review. Midweek, you only reopen the chat if something actually changed: illness, travel, a session you skipped, pain that's new. The rest of the week you follow the plan you already wrote.

This isn't medical advice and it isn't a diet. It will not find a deficiency in a CSV. If sleep will not recover, if mood drops and stays down, if weight or food is becoming a fight, that is a person, not a prompt. Use the model to keep the week honest. Use a clinician when the week isn't the problem.