SoterCare

BSc (Hons) Computer Science | Software Development Group Project

“SoterCare is an edge-first IoT and AI ecosystem that gives elderly individuals continuous, dignity-preserving monitoring, functioning like a virtual nurse by their side, 24/7. The system centers on a single discreet wearable, the Smart Thigh Node (ESP32-S3), fitted with a 6-axis IMU, IR temperature sensor, moisture clip, haptic motor, and SOS button. This streams data over BLE to an Edge Gateway (Raspberry Pi 5), which runs a custom on-device gait classification model, achieving 97.65% accuracy in detecting risky movement transitions before a fall occurs, triggering a silent haptic warning instead of a reactive alarm. Behind the gateway sits a NestJS microservice backend connected to three purpose-built AI agents. A real-time Safety Guard polls sensor data every 10 seconds and fires threshold-based alerts (falls, fever, hygiene anomalies) with near-zero latency, no LLM required. A Prescription RAG agent, built on PostgreSQL’s pgvector extension, lets caregivers ask medication questions answered strictly from that user’s own uploaded documents. A scheduled Clinical Report agent uses Gemini 2.5 Flash to generate structured, human-readable health summaries every six hours. Families and care professionals access everything through a React Native mobile app and Next.js web dashboard, with full functionality preserved even during internet outages thanks to the edge-first design. Where smartwatches and traditional PERS systems each cover one narrow function, SoterCare unifies fall prevention, hygiene detection, and AI medical reasoning into one private, camera-free system, validated by clinicians at National Hospital Kandy.”