sirmos/mamagemma

Maternal health assistant for community midwives in Nigeria — Gemma 4 Good Hackathon # 🤱 MamaGemma: Maternal Health Assistant for Community Midwives > Built for the **Gemma 4 Good Hackathon** by Kaggle × Google DeepMind · $200,000 Prize Pool --- ## The Problem In Nigeria, over **30,000 women die annually** from preventable pregnancy complications. Community health workers (CHWs) and midwives operating in primary health centres often face: - No reliable internet connectivity - No access to specialist consultation - Patients who speak limited English - Overwhelming patient loads with limited support tools A CHW in a rural Lagos clinic needs to make triage decisions in real time, without being able to "Google it." ## The Solution **MamaGemma** is an offline-capable AI assistant that supports community midwives during prenatal consultations. It: - Takes patient vitals, symptoms, and optional photos - Assesses danger signs using WHO/FMOH guidelines - Returns a structured clinical recommendation with urgency level - Summarises findings in **Nigerian Pidgin English** for the patient - Stores all records **locally**, no data leaves the device ## Why Gemma 4? | Feature | How MamaGemma uses it | |---|---| | **Multimodal input** | Accepts photos of rashes, wounds, medication labels, or ultrasound printouts | | **On-device / edge deployment** | Gemma 4 E4B runs via Ollama on a local laptop, zero internet required | | **Function calling** | Structured JSON output for consistent clinical formatting | | **Apache 2.0 license** | Free to deploy at PHC facilities with no licensing cost | | **Strong instruction-following** | Reliably follows WHO triage protocols in structured prompts | ## Demo ``` python app.py # Open localhost ``` ## Quick Start ```bash # 1. Clone git clone github.com cd mamagemma # 2. Install dependencies pip install -r requirements.txt # 3. Configure API key (get free key at aistudio.google.com) cp .env.example .env # Edit .env and add your GEMINI_API_KEY # 4. Run python app.py …