Hello i am build the Trash to Treasure is an offline AI-powered cleanup coach that helps users identify and properly dispose of waste using their phone or laptop camera
What I Built:
Explains the core loop: Go Outside to Spot Litter to Snap Real Photo to Local AI Classifies to Pick Up Safely to Log Real Impact to Pocket Phone & Keep Walking.
Highlights the dedicated Screen-Minimization Mode ("Touch Grass Mode") designed specifically to keep users off their screens and present in nature.
Highlights who it's for (hikers, ploggers, community cleanup groups, students, and eco-explorers).
Demo:
Step-by-step user walkthrough from opening the camera to real-time on-device classification, confirmation checkbox, and confetti impact summary.
Direct links to your live web app and GitHub repository.
Code & Architecture:
Details the Two-Layer Decoupling Architecture:
Layer 1 (Perception): Local Gemma 4 / Ollama vision model observing visual traits.
Layer 2 (Deterministic Rules): Local safety rules and municipal disposal profiles (preventing LLMs from hallucinating city recycling rules).
Highlights the strict zero-state initialization and cryptographic scan_id anti-duplicate point award mechanisms.
Why Does Open Innovation Matter?:
Zero Connectivity Required: Closed APIs fail in state parks and backcountry trails with 0 cellular bars.
100% On-Device Privacy: No personal photos or outdoor locations sent to corporate servers.
Zero Marginal Cost: Local inference on Ollama costs $0.00, removing financial barriers for volunteers.
Auditable Safety: Deterministic open rules prevent dangerous pickups (needles, batteries).
Prize Categories:
Use Gemma, Google's open-weight model, in building your project
GitHub Repository**: GitHub - Trash-to-Treasure
Top comments (0)