Nuba

The context: the 49% administrative burden

Primary care physicians in Spain operate under extreme workload pressures, frequently seeing 35 to 45 patients a day with less than ten minutes allocated per consultation.

Our field research and clinical time-use studies revealed a stark reality: physicians spend nearly 49% of their workday dealing with electronic health records and manual computer paperwork rather than looking at patients. In the consultation room, doctors sit behind cluttered desks with their backs turned, furiously typing symptoms into rigid, decades-old software.

This dynamic causes severe burnout among doctors and leaves patients feeling alienated. Compounding this, studies and patient surveys show that nearly 60% of patients forget their doctor's verbal instructions the moment they leave the clinic, causing anxiety, medication mistakes, and repeated appointments.

Nuba was born as a master's thesis project at UDIT to explore how ambient artificial intelligence could absorb administrative note-taking, returning human presence to medical consultations.

Field discovery and defining the service scope

We conducted deep interviews with primary care doctors and patients, combined with a quantitative survey of 83 clinic visitors.

The qualitative inquiry surfaced a crucial insight: while doctors wanted to understand emotional and social contexts—especially with psychosomatic symptoms where physical pain stems from underlying stress—the digital system only allowed rigid, chronological data logging.

Using the Double Diamond and Service Design frameworks, we defined the boundaries of the service:

• The core user: The primary care doctor. The tool had to function as a seamless background desktop plug-in, not an entirely new operating system.

• The passive beneficiary: The patient. They needed to leave with a plain-language summary of what was discussed, what their diagnosis meant, and their exact treatment steps.

Primary care service blueprint detailing in-room consultation interactions and ambient transcription layers.

Information architecture: tree testing with 60 physicians

A medical interface cannot afford ambiguity. Doctors navigate high cognitive loads and will reject any tool that requires a manual.

We mapped the information architecture across four core areas: patient history, live consultation recording, data trends, and settings. To validate navigation labels, we conducted two rounds of tree testing with 60 healthcare professionals using realistic task scenarios:

• Round 1: The initial test showed high success locating recordings (67%) and trend overviews (67%), but revealed terminology confusion. Doctors confused generic terms like "resources" with medical supplies, and audio volume calibration menus created unnecessary noise.

  

• Round 2: We simplified the tree down to four clean hubs: Mis Pacientes, Red de Datos, Grabación, and Configuración. Task completion rates jumped above 73% for ending recordings, and 83% for finding clinical trend analytics.

  

Through testing, we decided that microphone sensitivity and voice matching should happen via contextual tooltips and background calibration rather than separate technical menu items.

Designing Nuba Clip and testing AI prompt synthesis

A critical part of the project was designing the Nuba Clip: an automated A4 consultation summary generated at the end of the visit.

Because medical terminology is confusing and intimidating to everyday patients, the summary translates clinical jargon into concise, accessible language. It structures the visit into four predictable sections: what was discussed, prescribed medications, lifestyle recommendations, and next clinical steps.

To validate the feasibility of this synthesis, we simulated doctor-patient conversations and tested multiple prompt engineering iterations with language models. Early prompts generated long, intimidating academic summaries. Through iterative refinement, we calibrated the prompts to produce a compassionate, easy-to-read tri-fold layout that patients can keep on paper or access digitally.

Nuba Clip: An accessible, plain-language consultation summary translating clinical diagnoses into clear next steps.

Usability testing and service blueprinting

We built interactive high-fidelity prototypes in Figma and ran usability sessions with medical professionals across eight core workflows: starting ambient recordings, editing live transcription errors, searching historical patient notes, and dispatching summaries.

The results validated the service model: task success exceeded 90%, and participants praised the distraction-free UI. Testing showed doctors wanted inline text editing directly on the generated summary cards rather than hidden three-dot menus, which we incorporated into the final UI release.

We synthesized the entire system into a comprehensive Service Blueprint, detailing how ambient audio, real-time speech processing, local EHR encryption, and patient summary generation interact without requiring the doctor to look away from the patient.

High-fidelity desktop interface designed for distraction-free clinical use during consultations.