What if you could have an AI health coach trained on YOUR doctor?

One of the biggest changes to my practice over the last two years has been the incorporation of AI into the workflow of helping my patients achieve their goals.

When I look back at the plans I created five years ago when I started as an integrative medicine physician, they were thoughtful but relatively basic—they were also very labor intensive. I would spend many hours pulling together various data streams to build personalized plans for patients.

That changed 18 months ago as I began placing de-identified or anonymized cases into some of the LLMs to see how they would perform. As we now have experienced in various areas of our lives they perform remarkably well and are only getting better. They are not perfect—they are also not a total substitute for expert medical oversight—but overall I’ve been impressed by their ability to synthesize the multiple data streams, provide interpretations, and summarize it both for me as well as my patients. 

As I tested various AIs against each other, I saw that ChatGPT, Perplexity and Claude each had their idiosyncrasies. Over time, I have settled on using Claude for most of my analyses though I still occasionally test across platforms to ensure there are no ‘out there’ recommendations. That is my role as the physician using the tool. We are certainly in the stage of AI development where an AI can provide excellent results, but an AI plus a professional or expert overseeing the work is still likely to lead to the best outcomes.

As the medical science continues to evolve, it is good to have a second brain (in the case of my functional medicine practice, my own, informed by 30 years of practice in medicine) review it and provide commentary.

An AI Health Coach trained on my work

Over time, I have thought that the use of these models could be helpful as a virtual health coach. We have had exceptional coaches as part of our practice who I, and my patients, have found invaluable. There are, however, often multiple questions that come up between sessions or which can’t be fully covered in a session, with me or the health coach.

I thought: wouldn’t it be nice to pose a question to a virtual coach who would have all of my individual knowledge, learnings, and experience, as well as the database of knowledge and style of our practice?

Enter our virtual health coach.

How I built an integrative medicine AI health coach

I am not a coder. I never took a coding class in college nor have I ever dreamed I could create something using code. It is literally a foreign language to me. Java Script, python, SQL, fortran, cobalt (I’m dating myself) were words which have no meaning to me. When I read an article from a friend of mine, a similar coding novice, who coded something in Claude code for his business, I was intrigued.

I opened up the little box in Claude code and I typed in this query: Can I create a holistic health coach agent using the same database I created for my holistic health plans?

What came next wasn't exactly easy. But first, a little background.

The Virtual AI Health Coaching consumer industry is currently valued at $11 Billion (and projected to rise to $32 Billion by 2032). Here is a current list of applications:

• Weight & Metabolic

  • Noom — behavioral psychology + AI coaching, GLP-1 medical pathway available

  • Virta Health — diabetes/metabolic nutrition coaching

  • Lumen — breath-based metabolic coaching (fat vs. carb burning)

• Fitness & Movement

  • Hinge Health — AI-guided exercise therapy for pain/MSK

  • Fitbod — adaptive strength training based on past performance

  • Fitbit Premium — wearable-integrated AI readiness and recovery coaching

• Whole-Person / Lifestyle

  • Thrive AI Health — motivational interviewing AI, wearable sync, daily check-ins

  • Wellory — nutritionist-matched coaching with AI support

  • Noom (again, fits here too)

• Mental Health

  • Wysa — CBT-based AI chatbot, clinically validated

  • Youper — emotion/mood AI coaching

• Sleep

  • Sleep Reset — AI coaching based on chronotype and behavioral sleep data

• Women's Health

  • Flo — hormonal/cycle tracking with AI-personalized recommendations

Most of these are subscription-based ($10–$60/month range), direct-to-consumer, and app-native. 

The challenge is that none of these use comprehensive data sets of your personalized information tailored to deliver feedback based on an established plan. While it's true we can all upload personal information into any of the AIs and get some reasonable feedback, you will never be sure of the soundness of the science it is using.

You can prompt AI to use only specific information but for most will not know the reliability of the dataset and the output. One only has to spend a few minutes researching for the perfect nutritional regimen to be utterly confused on what to do. So, how do you tie in your personalized genetic markers, microbiome data, biometrics and lab values to create a truly knowledgeable agent that can provide specific information to assist me in my personal health care journey?

That set me down a path to create, after many hours of back and forth with Claude Code, a virtual health coach styled after my integrative medicine practice and personal philosophy of providing care.

It involved creating a database of information to query. Since I have written hundreds of blogs over the years I was able to upload that as information for it to ‘ping’ as it created the output. I asked Claude, separately with specific prompts, to create white papers for my practice based on Dignity Integrative Health and Wellness’ Four Pillars using the latest medical literature. I also uploaded summaries of scientific information on the latest research, how behavioral interviewing works in health coaching and various other specific information to be consistent with my style of practice. 

Examples of what to ask our AI health coach

To be clear, the intent is not to replace the actual coach working with you—the point is to provide an always-on, 24/7 extension of our existing care team, for whenever a patient wants to ask a question. Now, they can receive personalized care, consistent with our practice, whenever they want, for as long as they want.

While AI is very good it is not the same as another human being on the other end of the phone or video screen to help work through specific issues with you.

However, our new AI health coach is very good at answering questions like:

• Based on my genetics and lab values design a movement program using exercise bands that can help me achieve my goals

• Describe to me the impact my genetic information has on methylation and how that would impact my health.

• How would this specific diet impact my microbiome results?

• I’ll be traveling out of the country to Bhutan and need to try and adhere to my SIBO diet, what would you recommend?

All of these types of queries are readily answered by the AI agent. 

How it works

Each Dignity Integrative patient gets their own login to the virtual coach with a unique identifier so you can’t simply query other ones. From there you have a text box that you can ask whatever you want.

Previous queries are there as well. It also has built in safeguards and will not answer questions such as medication adjustments. Anything that triggers established parameters that I have set sends me a message to discuss that issue with me as the physician.

We welcome early feedback from our patients!

It’s not perfect, but over time I have taken feedback provided and refined the tool. One of the biggest challenges is how to incorporate updated information, lab and microbiome data to update the information. As I am finding, there is always more to do to make it better.

To date, I estimate I’ve spent 30 hours building in Claude Code using synthetic or de-identified data. Mostly it involved trial and error for things not working, copying code created into Claude code and rerunning the program to test. Trial and error. Trial and error over again.

The platform was built with patient privacy as a foundational requirement, including encrypted access, unique patient authentication, and no storage of identifiable health information in any third-party AI system.