My AI Personal Trainer, Nutritionist, and Health Advocate — How One Claude Project Became All Three

It started as a single conversation. It got long enough that Claude actually suggested I turn it into a Project — I didn’t even know that was a feature until it told me.

Quick explanation if you’re new to this: a Project in Claude is a dedicated space that remembers context across every conversation you have inside it, and lets you upload files it references going forward — so instead of starting from scratch every time, it actually builds on everything you’ve told it before. That’s the part that made this whole thing possible.

Once I had that, I saw there was a file section and thought, what do I even put in here? I started with my old blood work. That’s where it actually started growing into something real.

Now it’s with me constantly. Training, nutrition, recovery, even prepping for doctor visits. Here’s how.

Three rules, from day one

1. Don’t get hurt. Get hurt, you’re off the plan, and everything goes downhill from there — especially as I get older. Protecting my shoulder and knees matters more than any single workout.
2. Keep showing up. Some days I’ve got 45 minutes, some days I’ve got 10. Doesn’t matter — we adjust the plan to fit what I’ve actually got, but I still show up.
3. Tell the truth. It’s easier to be honest with an AI than with a person sometimes — no judgment, nobody talking behind your back. As an introvert, that mattered more than I expected.

There’s a line that’s become the philosophy underneath all three: every action is a vote for who you’re becoming. A missed workout or an off-plan meal still gets reported honestly, not hidden — because the honest report is itself a vote in the right direction, even when the behavior wasn’t. That’s changed how I recover from a bad day instead of spiraling over it.

Training

This isn’t just logging sets and reps — it’s pattern recognition across real data. Every morning it pulls my HRV, resting heart rate, and sleep from Apple Health before I even say anything, and uses it to calibrate the day: HRV above 100ms, push harder; under 60, back off, don’t grind through it.

The difference between this and a human trainer is memory. It doesn’t forget. It remembers exactly what I lifted last time and adjusts from there. When flat bench started bothering my shoulder at heavier weights, we didn’t push through it — we scrapped it for single-arm alternating presses. Better chest movement anyway. Rule one in action.

Nutrition

I talk to it, out loud, using talk-to-text. Tell it what I ate. But nothing gets trusted just because it’s been logged before — my bourbon chicken recipe went through a full re-weighing process (raw chicken, drained fat, rendered sauce, final container weight) because the number we’d been using was running about 9% high. That’s the standard now: verify, don’t assume.

I also wear a CGM (continuous glucose monitor), which turned “a walk after eating helps” from a guess into an actual proven pattern, with real numbers behind it.

It pushes back — it doesn’t just agree

When I wanted to blame a body-fat plateau on my six-day training split, it didn’t just nod along. It walked me through why my own CGM and HRV data actually argued against a stress-driven explanation, and pointed at a more likely cause — a thyroid conversion issue I’m now actively working through with my nurse practitioner. That’s the actual value: an honest second opinion built on my real data, not a yes-man.

Health data and doctor visits

I lean functional and natural health — my own experience is that conventional medicine is excellent when something’s acutely broken, but I haven’t found it especially useful for actually optimizing long-term health. That’s a personal view based on my own fifteen years of being told my numbers were “normal,” not a claim about medicine as a whole — but it’s why I work with a functional health nurse practitioner instead.

The medical side of this has turned into a real, ongoing investigation, not a one-time file upload. We’ve tracked eight years of my thyroid labs, and found a 2022 functional-medicine report — one I’d essentially forgotten about — that flagged the same T4 conversion issue years before anyone acted on it. That history became the actual case I brought back to my nurse practitioner, follow-up emails and lab comparisons included.

When I have telehealth visits, I keep voice-to-text running on a second device so it listens to the whole appointment and gives me a full recap afterward. Most people forget most of what they were told by the time they leave the room. I don’t, because it’s sitting there in context for the next follow-up.

To be clear: this didn’t replace my practitioner. It made me a far more informed, better-prepared patient walking into those conversations.

The honest limitation

It only knows what you tell it. No input, no insight. You have to actually keep talking to it, stay honest, keep checking in. Do that, and it stops feeling like a tool and starts feeling like a standing working relationship — an honest second opinion with a photographic memory, built on real data instead of guesses.

Where this led

This project is also the reason AI Coach Builder exists. Once I saw what this setup could actually do for me, I built a tool so other people could set up something similar for themselves — free, takes about 10 minutes.