NutriAI: A Partitioned Edge-Cloud AI Architecture for Privacy-Preserving and Personalized Mobile Health Management

The management of non-communicable diseases (NCDs) requires continuous nutritional monitoring. Existing mHealth solutions often suffer from a "dependency bottleneck"—relying on high-latency cloud processing or resource-constrained on-device models. We present NutriAI, an Android-based architecture that utilizes a partitioned AI strategy to optimize for latency, data privacy, and reasoning depth. NutriAI employs a dual-inference pipeline: on-device TensorFlow Lite for real-time perception and cloud-based Google Gemini API for complex semantic reasoning. A pilot study (n=20) and technical profiling demonstrate low latency, lean memory usage (~88.8 MB), and high user satisfaction. The hybrid edge-cloud design balances privacy and responsiveness, making it a viable model for personalized mHealth interventions.