Singapore's National University Hospital (NUH) deployed MedBot in July 2025 as an AI-powered virtual pharmacy assistant that has since delivered 28 man-hours of monthly labour savings and around SGD 15,400 in annualised cost avoidance, scaling from an initial 10 medications to AI-generated counselling content for 66 selected drugs across the Kent Ridge Wing Pharmacy and the satellite pharmacy at the National University Centre for Oral Health, Singapore.
"Every piece of AI-generated counselling content goes through a human-in-the-loop process — our pharmacists review and validate all content before it enters the database. This was non-negotiable," said Mr Jack Lim, Data Scientist at Kent Ridge Office of Innovation, NUH, summarising the governance model the team built. "The system delivers from this pre-approved database, not from live AI generation, giving us 100 per cent information consistency and eliminating hallucination risk entirely."
NUH pharmacists previously spent three to 20 minutes per patient on routine medication counselling across 2,500 prescriptions dispensed daily in outpatient and satellite pharmacies, leaving limited capacity for complex care. The multidisciplinary team built MedBot in a deliberate, phased rollout: clinical strategy was led by Ms Goh Sok Hiang, Principal Clinical Pharmacist and Clinical Lead for MedBot, with senior pharmacist Mr Wong Hong Yun driving the patient experience; Mr Lim steered the technical build. The first 10-medication launch validated safety guardrails (allergy acknowledgement, pharmacist review panels, escalation), after which the catalogue expanded only once strict safety milestones were met.
MedBot is built on the AWS Healthcare Commercial Cloud (HCC) and integrates with NUH's electronic medical records plus the Synapxe H-Cloud medication database — one of the first projects in the hospital to bridge its on-premise data centre with cloud in real time. Counselling sessions route through dedicated kiosks at the dispensing counter, with pharmacy typists identifying eligible patients at intake.
Over eight months of patient testing the team surveyed close to 1,300 patients; 98.7 per cent reported MedBot was easy to use, 98.9 per cent said the AI-generated information was clear, and 96 per cent felt comfortable proceeding with medications after using it. The team is now planning an interactive chatbot function and a mobile version of MedBot to take counselling away from the pharmacy counter, with the longer-term goal of moving from human-in-the-loop to human-over-the-loop as the system matures.