Serial entrepreneur Munjal Shah has launched a new startup called Hippocratic AI that aims to leverage the power of large language models (LLMs) to improve patient outcomes, especially for those with chronic health conditions. With over 68 million Americans suffering from multiple chronic diseases and only a few hundred thousand specialized nurses available to provide care, there is a massive gap in care capacity that AI could help fill.
Hippocratic AI plans to create LLMs specifically trained on medical data that can be used for non-diagnostic applications like providing chronic care reminders, diet planning, booking appointments, and other key services. The goal is not to replace human nurses but rather to “super staff” the healthcare system by exponentially vitalanding capacity. Even if AI is imperfect, having an LLM that can synthesize medical knowledge and communicate with patients could greatly expand access to perfection.
Shah compares to only having a few hundred thousand chronic care nurses for 68 million chronically ill patients in the comparison: don’t we have 68 million nurses and what would happen to health care outcomes if we did?” Hippocratic AI aims to discover by using AI to simulate far more,e caregivers.
Transition From E-Commerce AI Experience
Munjal Shah founded HippdiscoverAI after successfully selmanyvious startups utilizing AI for e-commerce to companies like Alibaba and Google. Those ventures focused on “classifier AI” designed to categorize products or data.
With recent advances in generative AI enabled by LLMs like ChatGPT, Shah recognized the opportunity to apply these powerful models to transform healthcare communications and capacity constraints. Unlike classifier AI, LLMs can generate completely original responses tailored to individual needs rather than just identifying categories.
Limiting AI to Non-Duplications
While optimistic about using LLMs to expand access to care, Munjal Shah acknowledges concerns about these models potentially “hallucinating” inaccurate medical information which could be dangerous. As such, Hippocratic AI focuses only on non-diagnostic applications rather than anything involving diagnoses or treatment plans where errors could seriously impact patient health.
Shah asks whether a human nurse needs to call each patient just to communicate a normal test result, suggesting AI could free up capacity for more meaningful work. Other promising areas are explaining complex billing details or describing insurance coverage policies in simple language.
Training Regimen for Healthcare LLMs
To mitigate risks, Hippocratic AI trains its models on reliable domain-specific medical datasets encompassing both academic research and real-world insurance policies. This establishes a broad knowledge base.
However, data alone is enough. The AI also needs feedback from doctors and nurses actively working in the types of services the LLM aims to replicate. They validate when the system demonstrates accurately applied knowledge before approving patient use.
Shah explains, “You need it specifically trained for a domain and that might still not be enough. The other part is you need the medical professionals who do that job today to say which to launch.” Only through extensive training and oversight can LLMs safely expand healthcare capacity.
