CHEW Copilot
Triage guidance that speaks the clinic's languages.
AI product · multilingual UX · 2025 · LLMs, multilingual UX, healthcare, triage
Source ↗Guidance for the thinnest tooling environment
Community health extension workers operate where the stakes are highest and the tooling is thinnest: brief consultations, mixed-literacy settings, and patients who may not share a language with the protocol documents. A generic chatbot demo fails here on three fronts at once — it assumes stable connectivity and long attention, it speaks one language confidently, and it optimizes for conversation rather than for a correct next action.
Guided steps, not open chat
Open-ended chat invites rambling precisely when a health worker has minutes. The assistant drives toward a structured outcome — symptoms in, triage guidance out — with each turn narrowing toward a decision instead of extending a conversation.
Five languages as a primary path
Nigerian frontline care crosses English, Yoruba, Igbo, Hausa, and Pidgin within a single clinic day. Language support is designed as a primary path with consistent clinical meaning across switches, and every response ends in an actionable recommendation — refer, monitor, or reassure — with reasoning visible enough to trust and short enough to read between patients.
Where it landed
The result is a focused concept for operational AI in public health rather than a generic chatbot demo — a working sketch of how language models can serve frontline workflows when the interface respects time pressure, language reality, and the need for auditable guidance.
The interface constraint taught more than the model did: advise, never decide, and show your working in seconds. Next would be offline-first behavior and field testing with real workers to see which escalation cues actually get followed.