NUHS Built Four Pharmacy AI Tools in Nine Months — the Data Plumbing Took Five Years
Source: GovInsider
Singapore's public healthcare system processes staggering volumes of pharmacy work every day — 730 medication reconciliations, 13,000 order reviews and counselling sessions across the National University Health System alone — and much of it eats up 173 pharmacist-hours on routine, low-complexity tasks.
Singapore's public healthcare system processes staggering volumes of pharmacy work every day — 730 medication reconciliations, 13,000 order reviews and counselling sessions across the National University Health System alone — and much of it eats up 173 pharmacist-hours on routine, low-complexity tasks. Four AI tools, built in just nine months by pharmacists themselves, are about to change that. But the real story is not the speed of the build — it is the five years of data infrastructure that made it possible.
The NUHS Cluster AI in Pharmacy (NCAIP) platform bundles four tools: MedTriage sorts patients at the pharmacy counter into those who can self-serve and those needing counselling; Admission and Discharge MedRecons catch gaps in medication lists as patients move across care settings; and MedVerify screens every order for dosing issues and drug interactions before reaching patients. In one validation test, MedVerify flagged a dangerous interaction between an anti-nausea medicine and a mental health drug — the pharmacist agreed and rejected the order. Once fully deployed, the tools are expected to reduce in-person dispensing time by up to 20%, saving as much as S$2 million.
The nine-month build timeline is deceptive. "If you look at any commercial solution now, they only have their own database," NUHS Deputy Group CTO for AI Dr James Lee told GovInsider. "The reason why NUHS could do it is because the plumbing work had already been committed five years ago." That plumbing is Endeavour AI, a shared data platform that centralises information scattered across the Epic national electronic health records and internal clinical systems, and Horus, a big data AI layer that orchestrates multiple models in a multi-agent setup. Without that foundation, Lee estimates the same tools would have taken "two to maybe even five years" to build elsewhere.
Equally important is how the tools were designed. Rather than IT handing down solutions, a core team of 26 pharmacists and interns across four NUHS hospitals drove the requirements. Heads of pharmacy from each institution worked through what project director Tan Chwee Huat called "frank and honest discussions" to prioritise the most impactful tools. Lee acknowledged that top-down AI deployments often fail because they "don't meet the users' needs" — NCAIP worked backwards from what pharmacists said they needed, then checked feasibility within a one-to-two-year window. All four tools are currently proof-of-concept stage; MedTriage has secured funding and targets deployment in 2027. Tan said discussions with national health tech agency Synapxe about extending NCAIP across other clusters have been ongoing for over a year.
Why it matters for Singapore: The NCAIP story is a case study in what separates AI demos from AI that changes workflows. Singapore's public healthcare system has been investing in shared data platforms for years — Endeavour AI and Horus are the kind of unglamorous infrastructure that never makes headlines but determines whether AI tools ship in nine months or five years. As the other two healthcare clusters watch this rollout, the question is whether Synapxe can replicate the "plumbing" at national scale, or whether each cluster will need to build its own.
