We're largely still working with LLMs at the moment -- Using them to immediately pull in relevant clinical information from previous encounters when a doctor sees a patient. Or using generative AI to edit doctors' messages to patients be more empathetic and... human (our pilot organizations have really loved this one so far). Using procedure codes on claims to guess if certain diagnoses were missed and to make more robust health risk profiles for populations as a whole -- these are a bit more NLP/data mining.