SORT/ED

The app icon
The friendly mascot

Inspiration

The lack of motivation to do a mundane and often inconvenient task of recycling. Inspired by apps like Strava, Forest and YeolPumTa, which were able to motivate masses to do the most reluctant-to-do tasks, we strive to achieve the same reach with SORT/ED and reach the same success for recycling.

What it does

SORT/ED is a mobile application that transforms recycling into a visible, rewarding, and social habit by leveraging gamification and AI. Instead of using guilt, the app motivates users through a friendship system where peers can see, like, and comment on each other's recycling logs. It features streaks, leaderboards, community challenges, and an environmental impact points system to make sustainability engaging for younger audiences.

How we built it

Frontend: A mobile app built using Expo, configured to capture photos directly through the camera and shrink them to 1024px for fast base64 upload (disabling the photo gallery to prevent resubmission fraud).

Backend Pipeline: An ngrok tunnel hands requests to a local Python server (main.py & agent_bridge.py), converting formats like iPhone HEIC to JPEG.

AI Orchestration: The photo is sent to Claude on Anthropic's servers along with functional tool descriptions. Claude pauses to consult a local rules database (sg_rules.py) for regional precision, then returns identified materials and confidence levels.

Scoring Engine: A custom Python script (scorer.py) calculates the precise SORT point output based on environmental impact data, assigning a unique scan_id to prevent user manipulation.

Challenges we ran into

Localised Knowledge Constraints: LLMs like Claude operate globally on remote servers and lack precise, localized disposal rules out of the box (e.g., Singapore's exact rules regarding regulated e-waste, ALBA bins, or battery fire risks in collection trucks).

Preventing System Exploitation: Creating a secure framework where users cannot game the scoring system required passing tool requests through a strictly controlled loop, ensuring the Python server, not the AI, calculates and records final point values.

Accomplishments that we're proud of

Separation of Logic: Successfully separating the AI brain from localized regulatory logic, meaning any adjustment to local waste rules requires a simple one-line edit in the local codebase rather than rebuilding prompt structures.

Frictionless UX with Safeguards: Implementing camera-only locks to ensure authenticity while generating automated prep guidance (like reminding users to empty a bottle or clear food scraps) to prevent bin contamination.

Scalable Business Framework: Designing an architecture capable of supporting B2B ESG goals through corporate-sponsored challenges (e.g., Coca-Cola) and faculty-level university campaigns (e.g., NUS) to drive organic growth.

What we learned

Building An App: We learnt how and what goes into the building of an app's frontend, backend and introducing AI into our app to improve the app's functionality.

What's next for SORT/ED

Corporate Sponsored Challenges: Launching monetization channels where prominent brands fund reward vouchers (e.g., unlock a $5 voucher for recycling 20 bottles) to hit their consumer-facing ESG benchmarks.

School & University Activations: Rolling out regional, campus-wide recycling competitions (such as inter-faculty challenges) to establish initial community networks, scale acquisition, and build consistent user loops.