What Changed
From Friction to Smoother Workflows
What we were aiming at was that a clinician could finish a task without stopping to look for a control. Here is how the shipped product differs from what came before it.
What we were aiming at was that a clinician could finish a task without stopping to look for a control. Here is how the shipped product differs from what came before it.
For the assistant, we looked at how people juggle several large language models and what the constant switching costs them. Subscriptions to a handful of separate AI services, time-consuming and confusing, always trading accuracy for speed or speed for cost. They wanted one place to get at the capability, with navigation they didn’t have to think about. For transcription, people complained about the same things in every tool: accuracy that varied, slow turnaround, too many steps just to upload audio, check a transcript and export it. They wanted fast, reliable, minimal cleanup, downloads easy to find.
.jpg)
Went through both modules as they stood and listed the places where users stalled: the response mode control, folder and prompt creation.
Looked at how clinicians and others handle multi-model AI and audio-to-text work, and at what made them abandon the tools they had.
Mapped the sidebar, chat panel, consultation view and note workflows at low fidelity, before any visual decisions were made.
Set the hierarchy that separates primary clinical actions from secondary utilities, so each module could change without disturbing the other.
Built the visual system for readability on data-heavy screens, with a restrained palette, clear structure and accent color reserved for the actions that matter.
Built both modules against the design system, with rendering that keeps transcription and note generation responsive while they finish.
