case study · Health · Hackathon winner · DevCenter × RumptyCloud Hackathon

Sabi

Maternal care, close to home.

Role
Product and full-stack engineering
Timeline
Hackathon sprint, 2026
Status
Live
Links
live ↗code ↗demo ↗
Stack
SvelteKitNestJSTypeORMPostgreSQLGroqYarnGPTAzure SpeechTailwind

A culturally rooted maternal health companion for Nigerian families, covering the weeks between antenatal visits: week-by-week education, Nigerian-food nutrition, symptom safety guidance and consent-based midwife links.

01

The problem

Maternal mortality in Nigeria is among the highest in the world, driven largely by delays in recognising danger signs, unmeasured self-medication with herbal mixtures, and gaps in care between sparse clinic visits. Mothers are also flooded with contradictory advice from relatives, vendors and social media.

Who it’s for

Pregnant women and their families in Nigeria, and the midwives and primary healthcare workers who look after them.

02

Goals & constraints

Goals

  • Calm, verified education for every week of pregnancy.
  • Nutrition advice built on Nigerian food (ugwu, egusi, moi-moi, ogi), not kale and salmon.
  • Symptom guidance that tells you clearly when to go to a clinic.
  • A care-team link the mother controls, so sharing is always her choice.

Constraints

  • It must never pose as a diagnostic tool.
  • Many users prefer listening, or speak Hausa, Igbo, Yorùbá or Pidgin.
  • Low data budgets: audio has to be downloadable once and replayed offline.

03

Approach

  1. 1.

    Four pillars: Learn, Nourish, Notice, Connect

    Week-by-week education, Nigerian-food nutrition and meal planning, symptom guidance with warning signs, and a midwife connection. Each one answers a specific gap between antenatal visits.

  2. 2.

    Evidence first, AI second

    The forty weekly guides are preprocessed with page citations and served by week, not regenerated by a model on each visit. Nutrient values come from the FAO/INFOODS West African Food Composition Table and published research on cooked Nigerian dishes.

  3. 3.

    Two portals, one consent model

    Mothers and midwives sign in through separate portals. A midwife only sees records after the mother generates a care-team code and approves the pairing.

04

Architecture

An npm workspace with independently deployable web and API apps, hosted on RumptyCloud.

  1. 01apps/web: SvelteKit portals for mothers and the care team
  2. 02apps/api: NestJS REST API, TypeORM entities per feature
  3. 03Rumpty managed PostgreSQL: pregnancy records, check-ins, meal plans, care links, messages
  4. 04weekly guide (English) → Groq translation → Hausa / Igbo / Yorùbá / Pidgin
  5. 05voice: YarnGPT → 9jaLingo fallback → Azure en-NG neural voice fallback → .mp3 download

05

Key decisions

Deterministic triage, not an AI diagnosis

Symptom guidance is a rule-based warning-sign check. For something this serious, predictable beats clever.

Keep source-backed numbers separate from catalogue data

The dish catalogue describes foods; only matching reference entries carry measured micronutrients. The app never passes off a catalogue item as a lab measurement.

A voice pipeline with three fallbacks

Indigenous-language audio is the feature most likely to fail, so it degrades through three providers instead of disappearing.

!

Challenges

Presenting with a dead laptop

The borrowed laptop died mid-call during judging. I finished the presentation from my phone, and we still took first place.

Trust without overclaiming

Writing copy that is reassuring but never says more than the evidence supports took as much care as the code.

✓

Outcome & lessons

  • 1st place at the DevCenter × RumptyCloud hackathon.
  • Live on RumptyCloud with demo accounts for judges.

What I learned

  • Cultural context is a product requirement, not a theme.
  • Honest limits make a health product more trustworthy, not less.

What’s next

  • Barcode lookup for packaged foods
  • Per-medicine source citations reviewed by a pharmacist
next projectPharmaCycle.AI→