2025 - 2026
AIoT Smart Eldercare
An AIoT smart eldercare solution for remote health monitoring and management.
About this project
An AIoT smart eldercare solution that integrates hardware device, real-time data analytics, and a mobile app/ website for caregivers. Built with React Native, Node.js, and PostgreSQL. Features analyzed predictions, real-time health monitoring, alert notifications, and secure data management.
Challenges & Solutions
Problem · WebSocket connections dropped frequently under poor mobile network conditions, causing missed health alerts.
Approach · Implemented reconnection logic with exponential back-off and a local offline queue that syncs on reconnect.
Result · Reduced alert delivery failures to near zero in simulated poor-connectivity tests.
Problem · Camera based HAR system misclassified certain activities due to lack of depth information, leading to false alerts.
Approach · Developed a 2.5D skeleton extraction model to provide depth cues for the HAR system, improving classification accuracy.
Result · Achieved around 90% accuracy in activity recognition in the testing environments.
Problem · The React Native app and web dashboard developed diverging UI patterns, increasing maintenance cost.
Approach · Extracted a shared design token set and component library used across both platforms.
Result · Halved UI bug count and significantly reduced cross-platform feature parity effort.
Key Techniques
WebSocket (real-time streaming)
Low-latency bidirectional communication is critical for health monitoring — HTTP polling would introduce unacceptable delays for alerts.
Designed the event schema and the reconnection protocol.
Go microservice
Go's concurrency model and minimal memory footprint make it ideal for handling thousands of simultaneous IoT sensor connections.
Built and optimised the data ingestion service from the ground up.
PostgreSQL with time-series indexing
Reliable ACID-compliant storage for health records with efficient range queries for trend analysis over days and weeks.
Designed the schema and indexing strategy for time-series health metrics.
React Native
Single codebase for iOS and Android caregiver apps with native performance and access to push notification APIs.
Led front-end architecture and implemented the real-time alert notification system.
Project Timeline
Discovery & Requirements
Researched existing eldercare solutions and potential multi-sensors integration solution and defined project scope
Architecture Design
Designed a microservices system with WebSocket for real-time streaming and PostgreSQL for time-series health data storage.
Hardware Integration
Integrated IoT sensors with the backend via a high-throughput Go service for concurrent device data ingestion.
Machine Learning Model Development
Developed 2.5D skeleton extraction and human action recognition (HAR) model to increase the accuracy of activity detection.
Mobile & Web Development
Built the React Native caregiver app and web dashboard with live health metrics, alerts, and patient history views.
Testing & Iteration
Conduct Testing and iterately upgrade the system based on performance metrics, i.e. accuracy, speed of alert delivery and stakeholder's feedbacks.
Outcomes
- Delivered a fully functional cross-platform system within the 12-month timeline.
- Impact-focused: Sustained 100% connection reliability across 20 concurrent streams with sub-1ms Redis pipeline latency, delivering 600 total FPS well under the 33ms frame budget.