Mama Salama — an offline Swahili companion for maternal & newborn safety
I lead Mama Salama, a Swahili, offline phone companion that helps a new mother check her own and her baby's danger signs in the weeks after birth. First place at Pitch II of the AI4Health Local AI Buildathon with 77.67 out of 100, and through to the Grand Finale on 12 November 2026.

The problem
In Tanzania the maternal mortality ratio is 104 deaths per 100,000 live births, and newborn mortality is 24 per 1,000 (TDHS-MIS 2022). Crucially, close to half of maternal deaths happen after the first day; in the days and weeks after birth, when the mother is home and out of reach of the health system. Postnatal-care coverage is low and most rural mothers never return for a check. At home, she often cannot tell whether bleeding, fever, a bad headache, or a baby who won't feed is normal or dangerous, and with about 70% of the population rural and far from emergency care, that delay can be fatal.
What we're building
Mama Salama ("Safe Mother") is a Swahili, offline phone companion. In the first three to four weeks after delivery, a mother uses it to check her own and her baby's danger signs, so a dangerous complication is caught early and she is guided to seek care in time.
The design is offline-first on purpose. Rural Tanzania has weak networks, power cuts, and costly data, but danger signs don't wait for a signal, often striking in the middle of the night. So the clinical logic and the Swahili language understanding live on the phone itself: it answers instantly, runs on a cheap device the family already owns, keeps her information private, and when a situation is unclear it errs toward "seek care."
My role & the team
I'm the team lead. Mama Salama is a deliberately mixed, two-university team of six: four data scientists from NM-AIST bringing the engineering, and two pharmacists from MUHAS keeping the clinical content grounded and correct.
Where it stands
First place at Pitch II, 23 October 2026, with an average of 77.67 out of 100 from six judges, top of the five teams going through to the Grand Finale. The finale is Thursday 12 November 2026 at the 6th Tanzania Digital Health and Innovation Forum, MUHAS.
A deterministic knowledge graph makes every decision. A 22.7M-parameter on-device model only reads the Swahili, and asks when it is unsure. 190 ms per clause, a 2.9 MB app, zero Android permissions, and no internet permission declared.
The clinical status, stated exactly: the content is reviewed and approved by the team's two MUHAS pharmacy members. It has not been validated by a practising midwife, and the app has not yet been used by a mother. Both are the work before the finale.
Evidence
- Buildathon announcement — LinkedIn
- Top-10 selection — LinkedIn
Updates
-
23 Oct 2026
Mama Salama placed first at Pitch II
An offline Swahili danger sign companion for new mothers, built for the AI4Health Local AI Buildathon 2026. Ten teams pitched, fiv...
-
04 Aug 2026
Nine principles, and the three I cannot answer for yet
Dr Felix Sukums took our AI4H buildathon cohort through the Principles for Digital Development. I ran our own project against all...