Update 2026: With the new technological improvements in AI, I’ve pushed a one-time update to refresh the model selection and tooling for the current AI ecosystem, though similar to before, I won’t continue on this project in the future as I have other things to work on.
In the first paragraphs of his book “Building a Second Brain,” Tiago Forte describes his journey caused by an illness of his throat that severely impacted his daily life to a point where basic human commodities were hard to endure.
As more and more doctors left him with inconclusive results, he started researching on his own, archiving received information into his own digital information system, searching for and eventually finding therapeutic techniques and practices that allowed him to alleviate his pain and claim back parts of his life.
Stories of chronic illnesses with indeterminate diagnoses are common and leave patients unsatisfied and without any proper treatment of the root cause. Meanwhile, more and more success stories on the internet show how LLMs can hint at diagnoses of chronic illnesses that eventually led to finding the cause or encouraged people to visit the ER or get urgent medical care when presented with a preliminary set of symptoms.
Recent improvements in large language models have advanced the capabilities in understanding medical text and also “diagnosing” or preliminary screening patients effectively before primary care.1
Despite these improvements, physicians have an innate advantage, that is having a way broader and in-depth understanding due to the prolonged medical history of ongoing screenings and preventive care that is all combined and accessible for the physician to consider when making a diagnosis.
Given this context, I have built an app centered around integrating medical data across categories into a little tracker. Additionally, a specialized large language model will then be enabled to chat and interact with the user’s history to provide more factually accurate and personal advice.2
Footnotes
-
I want to emphasize that I am not suggesting that LLM diagnoses should be considered to have medical validity; they can however, be helpful for obtaining a preliminary overview of potential causes, getting a second opinion, understanding differential diagnoses, and conducting your own research. In any case, a qualified physician should still be consulted. ↩
-
Since the latest update, I have replaced the Llama 2-based finetune with GPT-OSS-120B, which since has shown to be more effective in health-related benchmarks. ↩