How I AI: Building a Personal AI Wellness Coach to 'Feel 25 in a 40-Year-Old Body'

I was so excited to sit down with this week's guest, Lucas Werthein, Head of Technology at Cactus. His firm has done incredible work for clients like Apple, Coca-Cola, MTV, and even Beyoncé, but our conversation went somewhere much more personal.

As a competitive athlete, Lucas has spent years pushing his body to the limit, collecting a long list of injuries and surgeries in the process. He found himself with mountains of health data, but none of it was connected. He had blood tests, physical therapy notes, stats from his Whoop strap, nutrition plans, and MRI scans, all living in separate folders and apps. The real challenge wasn't getting the information; it was piecing it all together into a strategy that made sense.

So, Lucas showed me the exact workflow he used to build his own AI wellness coach inside ChatGPT. This is a tool he uses daily to manage his training, nutrition, and recovery, all with the goal of feeling 25 again in his 40-year-old body. He walked me through his entire GPT configuration, from the initial setup and data uploads to the specific prompts he uses for everything from navigating dinner parties to recovering from an injury. We also touched on two other really interesting ways he uses this same idea at work: creating 'synthetic clients' and an 'AI co-founder.'

Workflow 1: Building a Personal AI Wellness Coach

Lucas needed a way to pull all his complex health data together into one system that could give him clear, actionable advice. He wanted to create a 'performance strategist' that knew his body, his injury history, and the reality of his life as a busy executive and competitive tennis player. Here’s how he built it.

Step 1: Gathering and Uploading Diverse Data

First, he had to give the GPT a complete picture of his health. The really interesting part is the sheer variety of data formats he used. He didn't have to clean it up or standardize anything; he just uploaded it all.

His knowledge base included:

• Medical Imaging: X-rays of his left and right knee, plus MRI scans of his knee both pre- and post-surgery.
• Wearable Data: CSV files exported from his Whoop strap, containing detailed metrics on physiological cycles, daily journal entries (stress, sauna use, etc.), workout logs, strain scores, and sleep data (REM, deep sleep).
• Lab Results: Multiple blood exam reports from the past few years, uploaded as PDFs.
• Professional Plans: A detailed nutritional plan from his dietician and results from an InBody scan measuring body composition.

This really highlights how well modern AI models can process unstructured data. Lucas uploaded PDFs, CSVs, and image files in both English and Portuguese, and the model understood it all without any issues. That’s a huge relief for anyone who has felt overwhelmed by having their data scattered across different silos.

Step 2: Configuring the GPT's Persona and Goals

Once the data was loaded, Lucas wrote a detailed set of instructions to define the AI's role, objectives, and personality. This is the step that turns a general tool into his personalized coach.

• The Role: He instructed the model to act as his "performance strategist and health optimization coach."
• The Context: He specified that it should coach him like a "high-performance operator"—not a pro athlete, but someone balancing competitive sports, weightlifting, recovery, and the demands of running a company.
• The Main Objective: The primary goals were clear: "safeguard my joints, amplify my output, and extend my peak." He wanted to feel healthy and pain-free while performing like he did at 25.

I love the structure here. He gives the AI an input (its role and his data) and a clear desired output (how he wants to feel and perform). That kind of framing is key to getting relevant, useful responses.

Step 3: Defining Core Operating Principles and Hard Boundaries

To make sure the advice was practical and fit his personal philosophy, Lucas also defined a few key principles and things the coach should never do.

Four Pillars of Optimization:

1. Nutrition: Stick to his existing nutrition plan unless new data provides a compelling reason to change. Prioritize stable energy, low inflammation, and muscle retention.
2. Training & Load Management: Balance strength, endurance, and mobility, always prioritizing the protection of his joints. The coach must respect his Whoop readiness scores to prevent overtraining.
3. Recovery & Regeneration: Treat sleep, PT, mobility work, and sauna as non-negotiable parts of the training cycle.
4. Feedback Loops: All recommendations must be cross-validated against the data he provided. No random advice from the internet.

Hard Boundaries (What NOT To Do):

• No pushing through pain: Never recommend training hard when his Whoop data shows under-recovery.
• No unproven supplements: Avoid recommending trendy supplements or unscientific 'hacks.'
• No novelties: Stick to scientifically-backed methods.
• Act on red flags: If he reports soreness, low HRV, or poor sleep, the coach must prioritize recovery.

Step 4: Putting the Coach into Action

With everything set up, Lucas now uses the coach for daily guidance and bigger-picture planning.

Example 1: Navigating a Social Dinner

Facing a birthday dinner with lots of carbs and alcohol, Lucas asked for a plan:

"Good friends of ours gonna celebrate a birthday dinner, which means plenty of rice and saki. And so how should I manage my day to balance the fact that I'm gonna indulge in the evening?"

The AI coach responded with a precise, actionable plan: prioritize protein and minimal carbs for breakfast and lunch to prepare his body for the evening indulgence. He even sent a photo of his breakfast to the GPT, creating a positive feedback loop.

Example 2: Managing an Elbow Injury

While recovering from an elbow injury, Lucas used the GPT to augment the advice from his doctor and physical therapist.

• He fed it the official diagnosis and PT prescription.
• He provided daily updates on his pain levels, even uploading photos and videos where he pointed to the exact location of the discomfort.
• He asked a critical question: would he be able to play in an upcoming tournament on a specific date?

The coach took all this information and created a realistic recovery plan, complete with checkpoints for making decisions. The advice was the same as what his human experts told him, but it was presented in a way that was easier to digest and helped him manage the stress of the recovery process. It’s a great example of using AI not to replace experts, but to back up their advice and translate it into a format you can use day-to-day.

Workflow 2: Building 'Synthetic Clients'

Lucas doesn't just use this concept for his health; he also applies it in his professional life at Cactus. Getting feedback from busy, high-level clients can be a bottleneck, so his team came up with a clever solution: they create "synthetic clients."

Here’s the process:

1. Gather Public Data: They create a dedicated GPT for a key client.
2. Train the Model: They feed it publicly available information about that client—articles they've written, podcasts they've appeared on, public presentations, and other materials that reveal how they think and communicate.
3. Iterate Internally: The team can then query this synthetic client to get initial feedback on ideas. The goal is to get the work "to 80 or 90% of where we think she would agree with" before the actual client meeting.

Important Note: Lucas made it clear that they never use proprietary or confidential client information for this. It's all based on public data.

This helps them move faster, improve their work internally, and make every minute count when they finally get time with their real clients.

Workflow 3: Brainstorming with an 'AI Co-Founder'

In a remote or distributed work environment, you lose those spontaneous 'tap on the shoulder' brainstorming sessions. To fill that gap, Lucas created an 'AI co-founder.'

• The Setup: He created another GPT, this time loading it with information about his co-founder's thinking style, his own thought processes, and the context of specific business challenges they're facing.
• The Use Case: When he's wrestling with a thorny problem and needs a sounding board, he turns to his AI co-founder.
• The Outcome: It serves as an on-demand brainstorming partner, helping him avoid starting from a blank slate and work through ideas. He describes it as being "almost like business therapy."

It's a really smart way to simulate that collaborative thinking and keep creative ideas flowing, even when the team isn't in the same room.

The Future is Synthesized

What really connects all of Lucas's workflows is the idea of synthesis. Whether he's combining MRIs and sleep data, client interviews and articles, or a co-founder's way of thinking, he's using AI to turn an overwhelming amount of information into simple, personalized, and useful insights.

Lucas sees a future where this all happens automatically. He thinks that in five years, we'll all have a personal AI health coach. Our data will be collected in the background by tiny sensors and smart fabrics, and our personal AIs will talk directly with our doctors' AIs to create a completely connected healthcare experience. We'll look back at manually logging workouts and meals the same way we now look at typing GPS coordinates into a map app.

Ultimately, these workflows are about using AI to make better small decisions every day. They provide a clear path for achieving a goal, whether that's winning a tennis match, impressing a client, or just feeling your best. Talking with Lucas definitely inspired me to think about what kinds of AI coaches I could build for myself, and I hope it does the same for you.

Thanks to Our Sponsors

A huge thank you to our sponsors for making this episode possible:

• WorkOS—Make your app Enterprise Ready today
• Google Gemini—Your everyday AI assistant

Episode Links

• Watch on YouTube
• Listen on Spotify
• Listen on Apple Podcasts