TinyML Personal Health Companion

Inspiration

We were inspired by the need for continuous and personalized health monitoring, especially during heat waves, floods, pollution events, and situations with poor connectivity. Existing monitoring approaches may rely on generic thresholds or cloud connectivity, creating challenges for timely and privacy-preserving health support.

What it does

TinyML Personal Health Companion continuously monitors physiological and environmental parameters and detects potential health risks at an early stage. It provides personalized alerts, explainable AI recommendations, emergency support, live location, and nearest PHC recommendations while keeping core health detection available offline.

How we built it

We built the system using an ESP32-S3 with MLX90614, MAX30100, AD8232 ECG, GSR, BME688, and MPU6050 sensors. Sensor data is collected, preprocessed, and combined using sensor fusion. TinyML runs directly on the ESP32 for offline prediction, while a Personal Digital Health Twin learns individual health patterns. A Flutter dashboard displays health data and insights, with BLE and SQLite supporting offline operation.

Challenges we ran into

Our major challenge was achieving reliable health-risk prediction without continuous cloud connectivity. We also had to handle sensor noise, AI uncertainty, generic thresholds, environmental conditions, and connectivity loss during disasters. We addressed these using on-device TinyML, personal rolling baselines, sensor evidence validation, environmental correlation, and offline alerts.

Accomplishments that we're proud of

We developed an integrated system combining continuous health monitoring, TinyML edge AI, personalized health baselines, environmental intelligence, explainable alerts, and offline emergency support. We are especially proud of designing the system to preserve privacy while maintaining core health detection even when internet connectivity is unavailable.

What we learned

We learned how to integrate multiple physiological and environmental sensors, perform sensor fusion, develop an edge-AI workflow, and design personalized health-risk detection. We also learned the importance of privacy, explainability, offline functionality, and environmental context in healthcare systems.

What's next for TinyML Personal Health Companion

Our next goal is to improve the accuracy and robustness of the TinyML models, expand personalized health-risk prediction, strengthen environmental correlation, and improve emergency and PHC support. We also aim to further validate the system using real-world data and make the solution more reliable for disaster and everyday health-monitoring scenarios.