Slow and Steady

This project, TouchGrass AI, is an innovative micro-expedition application focused on minimizing screen time by integrating outdoor experiences into daily breaks. A key principle behind TouchGrass AI is ensuring the screen remains the shortest part of the user's experience. Users can set a personalized time budget ranging from 5 to 60 minutes and choose an environment such as a park, backyard, urban area, trail, or waterfront.

The app then generates a sensory outdoor micro-quest, which includes activities like observing nature, listening to birds, feeling tree bark, or exploring surroundings. Upon completion of the activity, users capture a single field photo and the data is recorded in a local SQLite database. Unlike traditional navigation and fitness apps, TouchGrass AI aims to keep the user engaged with the real world rather than further tethering them to their screens.

Built using open-weight AI locally, TouchGrass AI runs the MobileNetV4 ONNX model directly on the CPU, achieving inference in approximately 12 milliseconds. This design allows the app to operate seamlessly in areas with limited cellular connectivity, making it ideal for outdoor use. By processing field photos locally, TouchGrass AI ensures that users' personal surroundings are not transmitted to cloud-based services, preserving user privacy.

The project utilizes a stack that includes Python Flask, ONNX Runtime, NumPy, and Pillow for the backend, while the frontend is built with HTML5, CSS3, and Vanilla JavaScript.