MuindiKate/medtriage

Production-grade medical symptom triage API built with FastAPI, RAG architecture, and Claude AI. Designed for community health worker workflows in low-resource settings across sub-Saharan Africa. # MedTriage API 🏥 > AI-powered medical symptom triage for community health workers in sub-Saharan Africa. --- ## The Problem Kenya has **1 doctor per 10,000 people** — 10x below the WHO recommended ratio. The gap is filled by Community Health Workers (CHWs): minimally trained, working in rural health posts, making triage decisions with no decision support tools. The downstream effects are brutal. Serious conditions like sepsis and severe pneumonia get sent home. Minor conditions occupy scarce hospital beds. Both outcomes cost lives. Existing tools fail this context: | Tool | Why it fails | |---|---| | WebMD / Ada Health | Built for Western markets, requires reliable internet, not designed for CHW workflows | | eCHIS (Kenya's system) | Data collection only — no clinical decision support | | IBM Watson Health | Black box, enterprise pricing, no explainability | | Generic LLM prompting | Hallucination risk, no grounding in clinical guidelines | **The gap:** no affordable, explainable, API-first triage tool built for CHW workflows in low-resource settings. --- ## The Solution A **production-grade REST API** that CHW-facing applications integrate into. The CHW enters what they observe. The API returns what to do — and critically, *why*. ### Sample Request ```json { "patient": { "age": 4, "gender": "female", "weight_kg": 14 }, "vitals": { "temperature_c": 39.8, "heart_rate": 128, "respiratory_rate": 34, "oxygen_saturation": 94 }, "symptoms": ["fever", "difficulty breathing", "chest indrawing", "poor feeding"], "duration_days": 2, "history": ["no prior illness", "no vaccinations recorded"] } ``` --- ## What Makes This Different **1. Explainability as a first-class feature** Every response includes the full reasoning chain. In a clinical context, *why* matters as much as *what*. A CHW who can explain their referral decision is more likely to act on it. **2. RAG over hallucination** Responses are grounded in a curated medical knowledge base — WHO IMCI gu …