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

In this episode of How I AI, Lucas Werthein, Head of Technology at Cactus, shows the personal GPT he uses to organize training, recovery, nutrition, and health records around his goal of feeling 25 in a 40-year-old body.

Lucas had records from clinicians, blood tests, physical therapy, a Whoop strap, nutrition planning, body-composition measurements, and medical imaging. His problem was organizing that information for daily decisions.

The GPT helps Lucas summarize context and prepare questions, but it does not diagnose injuries, interpret imaging independently, or replace his doctor, physical therapist, or dietitian. He also demonstrates two workplace extensions: synthetic-client exercises and an AI co-founder for brainstorming.

Organize wellness context in a personal GPT

How to Build a Personal AI Wellness Coach Using ChatGPT: Build a private wellness assistant that organizes your records, professional plans, daily symptoms, and wearable trends into practical questions and routines without diagnosing, prescribing, or replacing qualified care.

Lucas configured a performance-strategist GPT around his existing clinical guidance, training history, work schedule, and goals.

Add the existing records and plans

Lucas uploaded the files he already had instead of first converting them into one format.

The reference set included:

• Medical records and imaging files from before and after knee surgery. The GPT stores this context, while qualified clinicians remain responsible for interpretation.
• Whoop exports with sleep, strain, workout, and journal data.
• Blood-test reports from several years.
• A dietitian's nutrition plan and an InBody body-composition report.

The files included PDFs, CSVs, and images in English and Portuguese. Lucas used the GPT to retrieve and summarize the collection, not to establish new medical findings.

Define the role and objective

Lucas wrote instructions that defined the GPT's role, the context it should use, and the outcomes it should prioritize.

• Role: performance strategist and health-optimization coach.
• Context: a company operator balancing competitive sports, strength work, and recovery.
• Objective: protect his joints, sustain performance, and avoid unnecessary setbacks.

The instructions connect the model to Lucas's own goals and professional plans rather than asking for generic wellness advice.

Set operating rules and boundaries

Lucas added rules for how the GPT should use his existing plans and when it should defer to recovery or professional care.

Four areas of attention:

1. Nutrition: begin with the dietitian's existing plan and do not change it casually.
2. Training load: balance strength, endurance, and mobility while protecting injured joints.
3. Recovery: include sleep, physical therapy, mobility work, and sauna in the schedule.
4. Context checks: compare a recommendation with the records and plans Lucas supplied.

Boundaries in Lucas's instructions:

• Do not recommend pushing through pain or training hard during poor recovery.
• Do not introduce unproven supplements or trend-driven hacks.
• Prefer methods already supported by his professional care plan.
• Treat soreness, low HRV, poor sleep, or worsening symptoms as reasons to slow down and consult the appropriate professional.

Use the GPT for daily planning

Lucas uses the GPT to organize questions and choices around meals, training, and recovery.

Example 1: Plan around a social dinner

Before a birthday dinner, Lucas asked how the evening fit with his existing nutrition 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 GPT suggested protein and fewer carbohydrates earlier in the day. That was a response within Lucas's personal context, not general nutrition advice for other readers.

Example 2: Track questions during elbow recovery

Lucas used the GPT while following care from his doctor and physical therapist.

• He added the clinician's diagnosis and physical-therapy prescription.
• He logged pain updates and showed where he felt discomfort.
• He asked the GPT to organize factors relevant to an upcoming tournament date.

Lucas reported that the resulting checkpoints were consistent with guidance from his doctor and physical therapist. Their advice remained authoritative, while the GPT helped him follow and discuss the plan day to day.

Simulate a client perspective with public material

How to Create a 'Synthetic Client' with a Custom GPT for Faster Project Iteration: Create a clearly labeled synthetic client from authorized public material to surface likely questions and priorities before a meeting, while keeping real client feedback authoritative.

At Cactus, the team creates synthetic-client exercises from public interviews, articles, talks, and presentations. These are internal hypotheses about a client's likely reaction, not the client's actual voice or approval.

The workflow has three steps:

1. Create a dedicated GPT for the exercise.
2. Configure it with public material that shows how the person communicates and discusses decisions. The team should still check whether its use is appropriate for the client relationship.
3. Use the simulation for early internal critique before presenting the work to the real client.

Lucas said the team excludes proprietary and confidential client information.

The simulation helps the team prepare, but only the client can confirm whether an idea fits.

Create an AI co-founder for brainstorming

How to Build an 'AI Co-Founder' in ChatGPT for On-Demand Brainstorming: Create an on-demand business sparring partner from documented company context and decision principles, using it to challenge a rough idea before bringing the stronger options to real teammates.

Lucas also configured a GPT as a brainstorming stand-in for moments when a colleague was unavailable.

• Lucas includes context about the co-founder's communication and the current business problem. A team adopting this pattern should limit the GPT to information and users authorized for that workspace.
• Lucas uses it as a sounding board when he needs to begin working through a difficult problem.
• It produces questions and possible angles so he does not begin from a blank page. Lucas compared the experience with business therapy.

The simulation supplements rather than replaces discussion with the real co-founder.

The shared pattern: organize context for repeated decisions

Across wellness, client work, and company brainstorming, Lucas packages scattered context so a model can retrieve it during a recurring decision.

Lucas expects more of this context to arrive automatically from connected systems. That is his view of a possible future, not a current claim that personal and clinical AI systems can safely exchange all health data.

Configure the GPT's Persona and Goals You are my wellness planning assistant. Help me organize dated information, follow my clinician, physical therapist, and registered dietitian plans, prepare questions, and choose low-risk daily actions. Do not diagnose, interpret imaging as a clinician, prescribe or change medication or supplements, or override professional guidance. Label facts, trends, and suggestions separately and state when evidence is missing.

Set Hard Boundaries (What Not to Do) If I report severe or rapidly worsening symptoms, breathing difficulty, chest pain, fainting, new neurological symptoms, serious injury, or another emergency sign, tell me to seek urgent local medical help. For new or worsening pain, stop the aggravating activity and help me summarize the timeline and questions for my clinician or physical therapist. Never reassure me solely from wearable data.

Lucas uses the assistant to organize questions and notice patterns across records, training, nutrition, and recovery. It does not diagnose an injury or replace his clinicians. The useful output is a better-prepared conversation with the professionals who already know his history.

The useful boundary is consistent across the examples: the GPT can organize context and questions, while people and qualified professionals remain responsible for consequential decisions.

Build a private wellness assistant that organizes your records, professional plans, daily symptoms, and wearable trends into practical questions and routines without diagnosing, prescribing, or replacing qualified care.