Offline clinical voice assistant

A talking textbook for healthcare workers.

AISHA answers clinical questions out loud, in Krio, from a medical library stored entirely on the device. It needs no internet and no grid power, and it is built with health workers in rural Sierra Leone.

The AISHA device, a white cylindrical speaker with the AISHA logo on the front
The setting

Built for offline clinics

Most frontline care in Sierra Leone happens at small rural facilities. Clinical guidelines exist for nearly every case these health workers see, but they are written in English, scattered across binders and PDFs, and hard to search in the middle of a shift. AISHA puts that guidance one spoken question away.

27× fewer health workers per capita than in high income countries
82% of health facilities have no institutionally provided internet (external source ↗)
A health worker at a desk in a rural clinic with an AISHA device beside her paperwork
AISHA in daily use at a rural health post outside Makeni.
How it works

Every answer comes from a real document

AISHA is a retrieval system. It matches each question against a curated library of clinical references stored on the device and reads back the relevant guidance. Nothing is improvised and nothing leaves the room.

Works off the grid

A full day on its internal battery and zero reliance on connectivity. AISHA works the same in a regional hospital and a village health post.

Speaks Krio

English guidelines, Krio conversations. AISHA bridges the two with speech models trained on Krio recorded across Sierra Leone.

Grounded in local guidance

The library is assembled with the Ministry of Health and partner facilities, adapting their protocols and standard operating procedures into a form that can be searched by voice.

Impact

Piloting in Makeni

Eight AISHA devices have been running in hospitals and Peripheral Health Units around Makeni since August 2025, in partnership with Sierra Leone’s Ministry of Health and Sanitation.

Health workers and patients around a table at a rural clinic with an AISHA device in the middle
A rural pilot site. Staff at this Peripheral Health Unit keep AISHA at hand to check guidance during the day.
A speaker presenting the AISHA device to a full hall of nursing students
Introducing the device to nursing students at Ernest Bai Koroma University, Makeni.
The AISHA team standing with officials from the Ministry of Health and Sanitation
Ongoing collaboration with the Ministry of Health and Sanitation.

Clinical evaluation

In November 2025 we ran a pre/post evaluation with ten healthcare workers at pilot sites. Each worked through realistic patient scenarios before and after a short AISHA learning module. Errors fell 35 percent on average, and 80 percent for recognition of hyperosmolar hyperglycemic state, a dangerous diabetic emergency that few had been formally trained on. It is an early result from a small group, and the evaluation continues as more devices go out.

8 pilot sites around Makeni
35% fewer clinical errors on average after using AISHA
80% fewer errors recognizing HHS, a serious diabetic emergency
Bar chart of clinical errors in recognition of diabetic emergencies for ten participants, before and after AISHA
Bar chart of clinical errors in recognition of HHS for ten participants, before and after AISHA

Errors per participant on scenario tests, before and after using AISHA. November 2025, n = 10.

Krio speech data

Krio has almost no public speech data, so we are building the dataset ourselves. Contributors across Sierra Leone record phrases through a WhatsApp platform we built, and native speakers review the transcriptions for accuracy. This data trains the recognition models that let AISHA understand questions the way people actually ask them.

400+ contributors across Sierra Leone
70 hrs of recorded Krio speech
35% human reviewed through our WhatsApp platform
Research

Peer-reviewed and presented

Designing an AI-Powered Voice Assistant Device for Healthcare Workers in Low-Resource Settings

Mazor, E., Leopold, M., Mundove, K., Earl, M., Kumaraja, J., Bangura, J., Mehta, K., & Obeysekare, E. Proceedings of the 2026 Design of Medical Devices Conference, Minneapolis, MN. April 20–22, 2026.

doi.org/10.1115/DMD2026-1051

Strengthening Last Mile Health Education Using Amazon Alexa-Enabled Devices: A Preliminary Case Study

S. Espinoza-Hernández, O. Hossain, Z. Urbaez, E. Yang, E. Ozturk, A. Thapa Kshetri. IEEE Global Humanitarian Technology Conference, 2024.

doi.org/10.1109/GHTC62424.2024.10771513
Team

Students and clinicians, working together

AISHA is developed by an interdisciplinary student team at Lehigh University working alongside clinicians and partners in Sierra Leone.

Seven AISHA team members holding assembled AISHA devices
Partners

On the ground in Sierra Leone

Ministry of Health and Sanitation, Sierra Leone National partner guiding adoption and scale across the health system
World Hope International Implementation partner supporting operations in Makeni
Makeni Regional Teaching Hospital Clinical partner and pilot site

Contact Us

For partnerships, pilots, research collaborations, or anything else, write to us.

ero324@lehigh.edu