My new personal trainer is a GEM (but Gemini is only half the story)
What happened when I turned a Gemini into a full-time nutritionist for two months
I’ve decided to share the (slightly fictionalized) story of a personal experiment I’ve been running lately.
A very simple experiment, but for me, it represents a turning point: it was the first real time I took GenAI out of my professional workflow and built something specifically for my personal life.
Before we dive in, let’s start with a couple of premises.
First, when it comes to health, AI should always be used with extreme caution. While AI is brilliant,
- it’s not a doctor
- it doesn’t have a medical degree
- it can sometimes “hallucinate” facts
- it doesn’t know your specific bloodwork or medical history
So, to put it simply, always consult a certified nutritionist for anything serious.
Second, my certified nutritionist.
Yes, I actually have one… and I trust her completely. I would never dream of replacing a human professional with AI. Why? Because a real nutritionist provides clinical oversight, empathy, and a deep understanding of human biology that a machine simply can’t replicate…
However, let’s be honest: following a meal plan is a nightmare.
It requires constant motivation and endless guessing.
Have you ever stood in your kitchen staring at a banana and an apple like they were complex calculus equations? “Which one prevents bloating? Which is better after pasta? Which one fuels my workout?” You want to make the right choice, but you’re standing there with a menu or a grocery bag, wondering if you’re about to ruin your whole week’s progress.
Since you probably only see your nutritionist once a month, you lack a direct line for those “emergency” questions. And don’t even get me started on traveling. Try following a diet in a foreign country where you can’t cook, the brands are different, and you can’t read a single word on the menu — unless, of course, you are perfectly fluent in reading ideograms or indian characters!
The “Walking” Breakthrough
The experiment didn’t actually start with a grand plan.
It started with a walk.
I was out for my daily stroll when I discovered that the Gemini App now allows you to have a continuous conversation without keeping your screen on. It felt like I had a person in my pocket. So, I started chatting…
It was the ultimate “getting to know you” session. While I walked, I told Gemini everything:
- I wanted to get back in shape and drop those few extra pounds.
- I shared my (sometimes questionable) eating habits.
- I listed my likes, my dislikes, and my workout routine.
- I emphasized that I needed flexibility because, well, life happens.
Gemini was as kind and understanding as only GenAI can be, but it was also surprisingly… sharp!
It didn’t just give me generic advice; it tailored every response to the habits I had just shared. It was looking for solutions that actually fit my life.
That’s when the lightbulb went off.
Why not hire this bot full-time?
Let’s be clear once again: I wasn’t looking for medical advice. I didn’t need a doctor in my pocket; I just needed someone to suggest recipes, meal pairings, and quick fixes for my daily life.
Building the “GEM”
For those of you who don’t know what a GEM is, in a nutshell: a GEM is a custom version of Gemini. You can create different “versions” of the AI, each with its own personality, specific knowledge, and set of instructions. It’s like having a team of specialists: one for coding, one for writing, and in my case, one for my fridge and my fitness routine.
I decided to use one of the golden rules of prompt design: Ask the AI to help you talk to the AI.
I told Gemini: “Based on our conversation today, write the instructions to create a custom GEM for a personal nutritionist.”
Get Riccardo Mestrone’s stories in your inbox
Join Medium for free to get updates from this writer.
Gemini delivered a detailed set of instructions. I tweaked them here and there to make the tone exactly how I wanted it, but the foundation was solid.
1. The Persona (the “Who”)
I told Gemini: “You are a highly qualified Nutritionist and Personal Trainer with a practical, sustainable approach. You are empathetic, non-judgmental, and focused on lasting health, not ‘crash’ diets.”
2. The knowledge base (the “What”)
This is where I defined the boundaries. I organized it into clear blocks:
- The Goals: A specific weight loss target per week
- The Pantry: I listed the local supermarkets I visit and the foods I actually like (lean proteins, specific vegetables, complex carbs).
- The Gym: I defined my routine — how many hours I walk and what equipment I have at home (like dumbbells).
- The Reality Check: I programmed it to know that “cheat meals” happen. I told it: “I might have a pizza or a beer. When I do, don’t tell me to skip dinner to compensate. Instead, help me balance my next meal.”
3. The Guiding Principles (the “How”)
I set “rules of engagement” for how Gemini should talk to me:
- Clarity over jargon: No medical fluff. Just tell me what to eat.
- Proactivity: If I ask about a snack, the GEM should remind me how it impacts my satiety or energy for my next workout.
- No pleasantries: I told it to be concise. I don’t need a “Hello, how are you?” every time — I just need to know if I should eat the apple or the banana!
3. The secret sauce
Then, I added the “secret sauce.” Over the years, I’ve collected a mountain of health material: PDFs from my nutritionist, books I read halfway, and guides I bought but never opened. It was a goldmine of information just sitting in my Google Drive.
(A quick tech note for the privacy-conscious: Since I’m using a Worskspace account, the material I upload remains private. It’s not used to train Google’s models, so my nutritionist’s advice and my personal guides stay between me and my GEM)
I uploaded all of it. I gave my new assistant all the professional advice and operative suggestions I had gathered over a decade.
The result? I created a personal trainer and nutritionist who knows my history, follows my philosophy, and answers in seconds — whether I’m in my kitchen or a restaurant in Delhi.
My Assistant at Work
Here is how our “friendship” looks in practice:
The Post-Workout hunger
Say I just got back from a 45-minute run. I might ask:
“I went for a 45-minute run today: what do you suggest for lunch?”
It doesn’t just give a generic suggestion. Because it “knows” me, it suggests things I might actually have in my kitchen, foods I enjoy, or things I can find easily. It might even tell me exactly where to buy a specific ingredient.
How does it know all this? Well… because I told it. During our initial setup and through our daily chats, I fed it my preferences, my shopping habits, and even what’s usually in my pantry. It’s not magic; it’s just a very attentive listener.
The “I’m Starving” Emergency
We all have those days where a small salad just won’t cut it. I can simply tell my GEM:
“I’m starving today: can you suggest a massive lunch that fits our goals?
And here is the best part: it doesn’t return an impossible or fancy proposal. Based on what you can realistically handle — whether you are at home with 10 minutes to cook or at the office — it proposes a consistent, high-volume solution. It knows how to “fill the plate” without breaking the rules we set.
The Social Dilemma
Sometimes, I just want to know how to navigate a social event without feeling like an outcast. I’ll ask:
“Is white wine or red wine better for my goals?”
(Spoiler: It usually suggests the one with fewer sugars or tells me to just enjoy one glass and drink extra water). It acts as a voice of reason — not a drill sergeant.
The “I Messed Up” Recovery
We’ve all been there.
“I couldn’t resist at lunch and had a massive plate of pasta, a 600g steak, and a huge gelato: what do you suggest for dinner?”
And no — skipping dinner is not the answer. Gemini knows me; it knows what I’m willing to do and what would just make me feel miserable. Et voilà! It suggests a reasonable, light, protein-focused meal to balance the blood sugar spike — a solution I can actually manage without difficulty.
The World Traveler
I’ve also found myself abroad, sitting in a restaurant and excited to try a new dish, with no particular preference other than wanting to stay aligned with my goals. Now, I just paste the menu or take a photo:
“I’m eating out tonight and this is the menu: what would you suggest?”
It’s like having a tiny expert living in my pocket, 24/7.
The Verdict
I have to confess: the results surprised even me.
Because yes, the results are there!
It’s clear that Gemini’s massive knowledge base, combined with the specific information I shared, allowed the model to give me advice that was not only accurate but, above all, sustainable.
But let’s be honest — was it just beacause of AI?
Is it because it gave me “magic” advice that a human expert couldn’t provide?
The truth is much simpler.
As I always tell my students, Generative AI should be viewed as a smart (junior) assistant. It can help us with repetitive tasks, or act as a “sixth sense,” and show us perspectives we might have missed on our own… but in the end, we are “the boss”, the ones who have to make the choices.
The solution was always in my own head, in my own willpower.
This experiment was a success because, subconsciously, I decided to commit to it.
I was the one paying attention to what I ate.
I was the one who chose to check the menu before eating everything in sight.
I was the one who skipped dessert when my GEM suggested I didn’t need it.
The fact that I need to cut back on sweets to lose weight?
I probably knew that long before I ever chatted with an AI.
Gemini didn’t give me “magic” powers — it just gave me the perfect tool to keep myself accountable.
And that, in itself, is already a great result.
