fokrulanthro16-eng/compassion-ai-sentinel

Offline clinical decision support AI for low-resource healthcare environments. # Compassion AI: The Offline Sentinel An offline-first Medical Decision Support Assistant for Community Health Workers (CHWs) in low-resource, rural African clinics — built for the Africa Deep Tech Challenge (ADTC) 2026. **This is not a chatbot.** It never claims to be a doctor, never diagnoses, and never invents medical facts. Every urgency decision comes from a deterministic, auditable rule engine; every piece of supporting evidence is a direct quote from a cited local guideline file; the LLM (optional, off by default) is only ever allowed to add a short educational note *after* the safety-critical decision has already been made and shown to the user. ## Project Overview Compassion AI runs entirely on a laptop with no internet connection, no cloud API, and no GPU. A CHW describes a patient's symptoms in plain English; the app returns a structured clinical assessment (a 5-level urgency rating, immediate actions, red-flag warning signs, and cited supporting evidence) in well under a second, then optionally generates a shareable Clinical Case Report the CHW can hand to a nurse or doctor. If a local AI model (via Ollama) is installed, it adds a short plain-language educational note on top of that decision — but the model is never in the decision-making path. ## Problem Statement Rural clinics and CHWs across much of Africa face three compounding constraints: **unreliable or absent internet connectivity**, **underpowered hardware** (often 8GB-RAM laptops with integrated graphics, sometimes running on solar or generator power), and a **shortage of trained clinical staff** to triage every incoming patient. Existing AI health tools generally assume constant connectivity, cloud GPUs, and forgiving power budgets — none of which hold in this setting. A tool that only works online, or only works fast on server-grade hardware, is not a tool CHWs in this context can actually use. ## Why Offline AI - **No connectivity dependency.** Every safety-critical decision (triage, …