# The Body Remembers but the Doctor Cannot- Experiments with AI in Personal Health > Source: > Published: 2026-08-05 04:01:19+00:00 Before we begin, I need to clarify this - As of today, AI still has an unresolved hallucination problem. Using AI in the context of health needs to be done very carefully and in conjunction with medical providers. Please do not accept any AI output blindly. With great power comes great responsibility for your own health. Ever since ChatGPT launched, I (like everyone else) have used AI for personal healthcare advice. LLMs were unreliable in their advice until December 2025. With GPT 5 and above, from my experience, the capabilities crossed the threshold from mostly wrong to mostly right. Especially for basic health questions. I have observed that reaching out to AI for quick personal healthcare queries is now getting more and more accepted within my social circles. And as the capabilities have grown, my reliance on AI for personal health advice has only grown in equal measure. My confidence grew with a combination of a) increased raw model intelligence and b) the various tools/workflows/skills that AI can use to power itself. The game changer for me was agentic harnesses such as Codex/Claude Code etc. There is a big gap in the quality of output a default out-of-the-box AI gives, versus what it can potentially do with the right tools. This is true for a whole raft of use cases, like coding etc., where the harness matters as much as the base AI model. And personal health is an area where this gap exists. I am building for that gap. I am also writing this post to jot down my experience so far. Over the last 2 years, my questions went from very basic questions such as “I have a fever, and I took 650 mg of paracetamol about Y hours ago. How long do I wait to repeat it?” to health problems that can only be attacked by talking with AI over a period of weeks/months and then gradually homing in on something that helps manage, understand or solve an issue. Issues like - chronic pain that has recurred for the last few years and gets worse randomly, new symptoms of continuous coughing (even though the person is not infected nor do they have any illness) that started randomly and now does not go away. These issues are complex, multifaceted and frequently do not have a clear answer. It is much more common to see cases where there are multiple health issues happening simultaneously that need to be managed or addressed. Doctors, of course, are the specialists that can solve these. In the current healthcare systems, a doctor is not available 24/7 to answer every tiny question one has. It is also not feasible for them to have the capacity to remember every detail about one person’s health and whatever affects it. Especially things that are from weeks/months/years ago. The health data is scattered across time and space, and it requires a tremendous amount of effort to keep track of everything manually. The body remembers but the doctor cannot. This is where AI excels. It is a health companion that can take the full context of an individual. I am talking about years of history that can be cross-checked, correlated, and matched to the latest research online and then tailored to that specific individual in that specific case for that specific query that is a pressing concern for that user at that moment in time. It is a superpower. This power, when harnessed correctly, can yield surprising insights into a person’s health that only a dedicated family doctor could uncover earlier. An involved family doctor is a very rare thing in the world. But an AI is not. ## The app for the gap It is called [CareLoop](https://careloop.karankurani.com/). It helps you manage your family’s health. The way it works is pretty simple on the surface. Pour all your data and questions into the agent. The agent will ingest it properly, create proper timelines, structures etc. to keep track of everything and then ask you follow-up questions. You tell it the concerns you want to explore and ask it to suggest next steps. Do those steps (with your doctor where it matters), and then review the result. Loop over and over again and see how it helps. It runs locally; no servers are required, and no data is shared with anyone externally (except the AI provider). All the data that the agent writes is stored locally within the working folder. The agentic app creates structured information out of all the data and then uses it to help the user in addressing their health-related queries. It has completely transformed the way I approach personal health. It makes me more informed about the state of my health. Second opinions are easy, and in-depth silly questions are frictionless to ask. It has also made me consider a lot of things I would not have thought of before. It will not solve problems that are unsolvable, but it will help you ensure that you are able to manage the complexity of health of your own body. Please note that it is alpha software that can make mistakes and should be used carefully and cross-checked with a doctor for any medical information. Monitor the AI chats diligently and audit the work. Verify important details yourself, and do not rely on it for urgent or high-risk decisions. Examples of things CareLoop has helped manage - - Person 1 - aging senior, tracking and solving chronic pain for long running shoulder and arm pain that has been going on for years and has been unsolved with multiple orthopaedic visits. Pain in the leg which flares randomly on and off and can get very bad at times. Blood Pressure issues, sugar values, muscle weakness from aging etc. How to manage all of it together along with a multitude of medications that come along with it. - Person 2 - person with a family history of cardiac history, family history of diabetes. The individual does not have any of it but is interested in preventive ways to tackle these so they don’t occur early to maintain a great quality of life till a good age. - Person 3 - managing of PMOS (formerly called PCOS), skin issues, random body aches, weakness, acid reflux issues, deficiencies of iron, dropping vitamin levels and much more. - Person 4 - sudden pain in the neck and arms along with tingling and pain. Navigating in understanding what is happening, understanding the treating doctor’s advice and POV (where the doctor does not have the time to explain) and then actually making an informed decision with their treatment. - and so on The set of things mentioned above is actually only a partial list of things for any given single person. In reality, there is much more information for each of the above. The actual health history that gets surfaced once we involve AI in the loop is fascinatingly much more detailed than what one might expect. This is true even for healthy individuals! In one session - a passing comment from a previous session (drinking tea at a particular time of the day) was highlighted as one of the possible causes that the user had never considered. In another session - the pain flare triggers investigation went into extreme detail about a person’s daily routine in a Q&A session that lasted for 1 and a half hours. In the end a simple low-risk action was proposed which, in fact, helped and solved the pain flares completely. A sample of questions from Person 2 above - - Even though my lipid panel is normal, is it normal enough for my case? Especially with family history? - When should I start considering preventive scans? What about the latest calcium CT scans that I hear on twitter? - How should I change my food intake on the weekends to have better control of sugar? is having yogurt with fruits in a protein bowl ok even though fruit is sugary? - I heard that cottage cheese (paneer) with spinach negates the protein intake because the body does not absorb it in that combination. - Here is my health data from Apple Watch - I am feeling more tired afternoons on some days and have energy crashes sometimes and it only started a couple of months ago. What could have caused it? This single person has had hundreds of their questions answered with their full context in mind instead of general health advice that does not always translate to the individual. The sheer amount of things that the AI needs to personalize to get into its context is very varied and different. Lab values, lifestyle factors, physical activity, health device data, data from images of prescriptions/reports, daily routines, food habits, social habits and more. There is another weakness in this tool apart from the hallucination and mistake checking. The medical community rightly states that a lot of things happening in our bodies are just quirky and do not indicate any underlying health issue, and it’s better to just live one’s life rather than look into anything. AI takes everything with equal seriousness and does not have a judgement that a human does. This means it can lead to overfixation on benign stuff, health anxiety etc. As a user, it is critical to understand the limitations of this tool and work within those boundaries. My hope is that as AI progresses, we see the same kind of impactful gains that the coding, design space, knowledge workers etc. have seen within their respective professions. This is just the beginning, [CareLoop](https://careloop.karankurani.com/) is built with things that were science fiction five years ago and the impact of AI on personal health will be huge over the next decade.