Alera: An Intelligent Habit-Tracking Platform Combining AI Coaching and ML Prediction

Alera is a full-featured habit-tracking app built with React Native, designed to break the limitation of traditional apps that only focus on recording. It integrates an AI conversational coach and machine learning prediction pipeline, providing personalized advice through data maturity stratification. Its tech stack covers mobile development, backend services, machine learning, and DevOps, showcasing the complete architecture of a modern full-stack application.

The design was inspired by reflections on traditional habit-tracking apps: users often face two major issues—lack of effective data interpretation and lack of continuous motivational guidance. Through AI coaching and predictive analysis, the project transforms passive recording into active intervention, making data serve behavioral change. The development approach combines academic research and engineering practice to build a comprehensive habit-building platform.

Technical Architecture: Frontend uses Expo + React Native + TypeScript + NativeWind; backend is designed with Supabase first—integrating Auth authentication (email/OTP verification), PostgreSQL database (RLS security policies), edge functions (metric calculation/AI conversations), and OpenAI API for AI capabilities, reducing integration complexity and improving efficiency.

Dual-Mode Design:

• Numeric habits: Suitable for quantifiable scenarios (water intake/reading pages, etc.), tracking progress and completion rate;
• Binary habits: Suitable for completion/non-completion judgments (meditation/medication), lowering the threshold for recording; Both types support daily/weekly/monthly target frequencies.

The AI conversational coach is a core feature, enabling personalized interactions based on user behavior data, habit history, current status, and past chat records. It integrates context (user profile/habit progress/metric data/chat history) via Supabase Edge Functions, answering progress queries, providing encouraging reminders, and even proactively intervening when users might give up—lowering the threshold for data understanding and making habit-building more human-centered.

ML Prediction Pipeline: Located in the ml/ directory, implemented with Python + scikit-learn, updating prediction data daily at UTC 6:30 via GitHub Actions.

Data Maturity Stratification:

• Locked layer (<14 days): Insufficient data, prediction functions locked;
• Basic layer (14-29 days): Provides continuous day risk assessment and habit trajectory analysis;
• Complete layer (≥30 days): Unlocks target completion time estimation and optimal reminder time recommendation; Prediction types include continuous day risk, habit trajectory trend, target completion time, and optimal reminder period.

Cross-Device Support: Habit records can come from the mobile app or Apple Watch; the system logs the source device to enable seamless cross-device tracking.

Supervision Mode: Users generate a unique supervision token to share with trusted individuals (family/coaches, etc.). Supervisors link their accounts to help create and manage habits, while the supervised user retains recording permissions—suitable for scenarios requiring external accountability, such as rehabilitation training or study supervision.

Alera demonstrates the complete form of a modern habit-tracking app: from basic recording to intelligent AI conversations, from metric display to ML predictive analysis, its technical architecture provides a reference for similar applications. In the future, we will improve the Apple Watch companion app and optimize UI/UX. Its AI+ML dual-drive model may represent the future direction of personal health management apps.