Credit: Me, Google Hardware Product Sprint'23, Google Taiwan
What I've Done:
- Attained 95% accuracy in autonomous waste identification by few-shot learning with domain-specific image data
- Demonstrated <150ms low inference latency with a computation-efficient LiteRT (former TFLite) Flask backend on Raspberry Pi
- Collaborated with engineers (hardware, software), designers (UI/UX, ID) and PM to create immersive educational experience
- Delivered award-winning product recognized as Best Product of the Year and Product for Social Good by Google Taiwan
Motivation: Boosting Environmental Education Experience with Technology
Environmental sustainability education for young students is crucial for building a greener future, but traditional
recycling education often lacks engagement and interactivity. Students need hands-on, interactive experiences
that make environmental responsibility feel rewarding and fun, rather than burdensome or abstract.
Our objective was to create an intelligent recycling system that combines AI-driven waste classification with
gamified educational experiences, making environmental learning both effective and engaging for students.
Our Product: GREENIFY--Your Recycle Mentor
GREENIFY represents a breakthrough in environmental education technology, combining computer vision, embedded ML, and interactive storytelling to create an engaging recycling experience. The system uses real-time waste classification to provide immediate feedback to students, while immersive storylines adapt based on user choices, creating personalized educational journeys that encourage better environmental habits.
Final Pitch @ Google HPS'23
The core innovation lies in the synergy of efficient AI-powered waste recognition and educational gamification. The interactive, physical interface provides immediate feedback and educational content, while the system's adaptive storytelling feature ensures that each interaction is unique, maintaining student engagement through personalized learning experiences . As an ML Engineer in our team, I developed embedded software and optimized on-device ML inference that recognizes 10 kinds of waste materials using EfficientNet, TensorFlow, and LiteRT within the Google ecosystem.
During the 2023 Google Hardware Product Sprint, our team successfully demonstrated how technology can transform environmental education. The project's recognition as both "Best Product of the Year" and "Product for Social Good" highlights its potential to make a meaningful impact on sustainability education while showcasing the power of interdisciplinary collaboration in creating innovative solutions.
Takeaways
Throughout this experience, where people from different fields worked together and strived to make a meaningful impact in education and environmental sustainability, I believed that the synergy of disciplines is what made the project successful. I will keep seeking opportunities to work interdisciplinarily to create a better environment, a better world.
