An artificial intelligence system developed in southern Brazil is screening medication orders across public hospitals in the Unified Health System (SUS) to stop drug errors before they reach patients. The tool, developed by health-tech non-profit NoHarm, drew national attention this week when physician and TV host Thelma Assis signed on as an ambassador to back the initiative, according to a report by Folha de S.Paulo. The software acts much like an automated spellchecker for clinical pharmacology: rather than catching typos, it cross-checks lab tests, patient charts, and chemical profiles to flag when two drugs clash or when a dosage exceeds safe medical thresholds.

Under the hood, the system relies on natural language processing to read unstructured clinical notes. Think of it as a silent co-pilot assisting hospital pharmacists during grueling shifts. When a physician prescribes an antibiotic based on body weight for a child, or orders a sedative for someone with compromised kidney function, the algorithm scans raw inputs recorded in hospital management software. In federal teaching hospitals run by Ebserh, such as the Hospital Escola at the Federal University of Pelotas (HE-UFPel), the system connects directly with the public network's AGHU software. In Pelotas alone, the tool checked 9,484 prescriptions and 130.337 medication orders, with attending doctors accepting roughly 70% of the pharmacists' AI-flagged recommendations.

NoHarm finances its mission through a cross-subsidization business model. The platform is offered entirely free of charge to public beds in the SUS, while charging commercial subscription fees to private hospital chains such as Rede Mater Dei. For the startup, co-founded by pharmacist Ana Helena Ulbrich (who was listed among the world's most influential figures in AI by Time magazine), the arrangement creates a practical growth loop. The public health sector provides vast volumes of anonymized clinical records to validate and refine the software, while private contracts fund engineering and payroll. The team reports having analyzed more than 175 million prescriptions across 200 healthcare units nationwide.

What we do not know yet is how far this software can scale beyond well-equipped university hospitals. Machine learning models require structured, digitized health records to function. In smaller municipalities across Brazil, primary care centers still rely on paper files or legacy databases that cannot communicate with third-party software. Furthermore, the algorithm does not change a single prescription on its own; it merely arranges incoming charts into a prioritized risk queue. Without enough trained pharmacists on staff to review notifications and talk to prescribing doctors, algorithm alerts risk becoming digital noise that hospital teams simply ignore.

For observers tracking digital transformation in public healthcare, the key metric to watch in the coming months is whether federal and municipal health authorities invest in modernizing electronic health records to make integration possible. Upgrading medical software is not just an administrative chore, but the foundational step required for algorithmic checks to catch prescription mistakes before drugs are administered at the bedside.