MedSync AI
Structured AI support for medication reconciliation — without replacing clinical judgment
Private clinic pilot completed · Reduced reconciliation time per admission · Improved flagging consistency
The Problem
Medication reconciliation on admission is one of the most error-prone, labour-intensive tasks in a busy clinic. A patient with multiple chronic conditions can take 30–45 minutes to reconcile manually — cross-checking reported medications against the formulary, identifying duplicates, and catching potential interactions. The process is done under time pressure by staff who may not have a pharmacy background.
What We Built
MedSync presents a structured intake form that guides staff through collecting a complete medication history. The data is cross-referenced against the clinic's formulary database, and an LLM layer drafts a structured reconciliation summary — grouping medications by therapeutic class, flagging items that need pharmacist review, and surfacing potential overlaps for clinical assessment. The output is explicitly a working draft: everything is labelled for review, nothing is presented as a final clinical decision. A pharmacist-reviewed flagging framework underpins what the model is prompted to surface.
The Outcome
Measurably reduced per-patient reconciliation time during pilot and improved documentation consistency across admissions staff
What We Learned
Explainability matters more than capability in clinical tools. The first version surfaced flags with no rationale — nursing staff ignored them. Once we restructured the output to show the specific reason each item was flagged, in plain language, adoption changed. Clinical users don't need a confident AI; they need a transparent one.