The original idea
Improve the writing
Take rough notes and return a clearer draft without changing the clinical meaning.
Founder story
The original problem was still the right problem.
I built NurseNote because nurses often know exactly what happened yet still lose time turning hurried shorthand into a clear record. The product took a detour. The evidence brought it back.
The detour
Teach through cases
NurseNote became a small assessment product. Three cases were too little value for the organisational price being tested.
The return
Keep the job, strengthen the boundary
The note improver is back with de-identification gates, constrained output and visible human checks.
What changed in my thinking
The assessment direction sounded safer and more defensible, but the product value was too indirect. A buyer was being asked to imagine an eventual library, reporting layer and validated programme that did not yet exist.
Note rewriting has the opposite advantage: the user sees the outcome within seconds. The danger is overclaiming what AI can do and ignoring information governance. So the answer was not to abandon the job. It was to design the limits into the interaction.
NurseNote now asks for fictional or genuinely de-identified text, blocks common identifiers twice, refuses clinical decision-making and separates the draft from what the nurse must verify.
The test is simple now.
Does NurseNote save enough effort, often enough, while keeping the meaning and responsibility with the nurse?
Try it with a fictional example