MehGerald/MamaAlert

Automated maternal danger signs screening via scheduled voice call in local Nigeria language with real time WHO triage and instant health- worker alert for high for high risk cases. Build as working end-to-end simulation for wema bank hackathon 7.0 # MamaAlert **Automated maternal danger-sign screening, triage, and health-worker alerting for antenatal patients without smartphone access.** MamaAlert calls registered pregnant women on a schedule, asks six WHO-based danger-sign questions in their preferred language, scores each response, and immediately alerts a health worker when a response is dangerous — no smartphone, app, or literacy required on the patient's side. This repository is a **working, end-to-end simulation** of the full system: real scheduling logic, a real scoring engine, a real event pipeline, and a live console — with the telephony, text-to-speech, and speech-to-text layers mocked so it runs anywhere with zero cost and no API keys. Every mocked component is built behind the exact interface its real counterpart would use, documented inline, so moving to production is a matter of swapping implementations, not redesigning the system. --- ## What this demonstrates Press one button and watch it happen in real time: - **7:00 PM, both facilities, automatically.** Primary Health Center A and Clinic B each have 10 registered mothers. At their scheduled call time, every mother is called — no one dials a number by hand. - **Concurrent, not sequential.** Both facilities' rounds run at the same time, and each facility dials multiple mothers in parallel, the way a real call round would. - **Per-mother language and routing.** Each mother is called in her own preferred language (English, Yoruba, Igbo, or Hausa). Clinic B's local-language calls route over an MTN SIP trunk; PHC A's routes over Africa's Talking's Voice API — modeled as two live options from the workflow doc, selectable per facility. - **Real IVR flow.** Opening line → six danger-sign questions → closing line, with a simulated spoken response, a transcription confidence score, and a DTMF (keypad) fallback when confidence is low — mirroring exactly how the real call needs to behave for a rural network connection. - **Live WHO-based scoring.* …