Lowfire
An experiment in AI as a clinical second opinion
Most EHRs surface a critical lab abnormality and stop there. The clinician still has to recognize the pattern, recall the guideline, and translate it into orders, often under time pressure. Lowfire is a small experiment in what comes after the alert: a deterministic algorithm produces the orders, an AI voices the recommendation as a colleague would, and the clinician decides.
Why this matters
Chemotherapy works by killing cells that divide quickly. That includes cancer cells, but also neutrophils — the white blood cells responsible for mounting the body's inflammatory response to infection. When the neutrophil count drops below a threshold, the patient enters a window where their immune system can't generate the normal signals of infection.
The urine looks clean even when there's a urinary infection. The chest X-ray looks normal even when there's pneumonia. Vital signs may hold even as bacteremia progresses. Fever is often the only signal left. That's why a low ANC plus a fever is one of the most actionable lab combinations in medicine — the empiric antibiotic window is measured in minutes, not hours.
All drug names redacted to class level. No real orders placed. No clinical use. The deterministic algorithm encodes IDSA / ASCO neutropenic fever guidelines.
Loads a worked example. You will walk through the patient modifiers as the algorithm composes the case.