KephothoM/SihaLink

AI-powered disease surveillance swarm for Kenya's community health system. Detects outbreaks, silent pandemics, and cross-county spread in real time — built on Google ADK, MongoDB Atlas, and Gemini. # SihaLink 🏥 ### AI-powered disease surveillance swarm for Kenya's Community Health Volunteers --- _Built for the Google Cloud Rapid Agent Hackathon — MongoDB Track_ --- ## The Problem Every year, disease outbreaks in Kenya kill people who could have been saved — not because the healthcare system lacks knowledge, but because it lacks **speed**. A CHW (Community Health Volunteer) in a remote Kisumu ward notices a cluster of cholera cases. She files a paper report. It travels by hand to the sub-county office. By the time the data reaches Nairobi, three weeks have passed and the outbreak has spread to four other counties. **SihaLink breaks that chain.** --- ## What It Does SihaLink is a **multi-agent AI swarm** that gives every one of Kenya's 100,000+ Community Health Volunteers a direct line to an intelligent outbreak detection system — through their phone, in their own language. A CHW speaks a patient report in Dholuo. Within 30 seconds: 1. **Intake Agent** — transcribes and clinically extracts the report (syndrome, triage, symptoms, vitals) using Gemini 2.5 Flash 2. **Geo Agent** — maps the encounter to ward/sub-county/county and finds the three nearest facilities with real drive times 3. **Data Agent** — stores it in MongoDB Atlas with a Voyage AI vector embedding for semantic search 4. **Surveillance Agent** — checks if this is part of a growing cluster against 4-week rolling baselines across all 47 counties 5. **Notify Agent** — fires a Telegram message to the facility: _"Incoming RED — 6yr, severe dehydration, ETA 22 min"_ 6. **Contact Tracing Agent** — maps who else the patient may have exposed, assigns CHW visits All autonomous. Human confirmation only for RED/YELLOW referral dispatch. --- ## Architecture ``` ╔══════════════════════════════════════════════════════════════╗ ║ SihaLink Swarm ║ ║ ║ ║ ┌──────────────────┐ HTTPS …