docwise · ai scribe

context

Docwise is the company; the product is an ambient AI scribe for NHS doctors. The scribe has to disappear during a consultation and then return with a clean, structured document that a clinician can sign off on without rewriting.

I worked under the CTO across the live product and the marketing surface, shipping 178 commits across ten weeks.

what i built

Live audio capture via WebSocket and AudioWorklet, with an FFT-driven waveform so the clinician sees that the room is being heard without the UI screaming for attention.

Parallel document generation so the wait does not stack linearly. Role-aware UI across three distinct clinician roles: GP, Consultant, Junior Doctor. Each role surfaces a different shape of document and a different cognitive load.

Marketing site as the second front: ten routes, 85+ components, ready for clinical buyers and ICB stakeholders.

engineering choices

AudioWorklet was non-negotiable; the main thread cannot stutter during a real consultation. WebSocket framing was tuned so partial loss never produces a corrupted document.

Role-aware UI was modeled as composition over branching, so each clinician role is a thin shell over a shared, well-tested core.

outcome

The scribe is live with a small cohort of NHS clinicians and the marketing site is the public face of the product going into 2026.