Inspiration
Most people want to recycle correctly, but the rules are confusing and they change by item condition. A glass jar is recyclable until it is broken, a cardboard box is recyclable until it is greasy, and an aerosol can is recyclable until it still sprays. SortRight is the recyclable-or-not sorter for the items people get wrong.
Theme Track — Rethinking Waste: The waste problem we targeted — and how we'd measure success
Contamination and wishcycling are the core problem: hopeful items tossed into recycling contaminate good loads, and hazardous items such as batteries can cause fires. Many items are condition-dependent, so a flat item-to-bin table gives the wrong answer. We would measure success by (1) verdict accuracy against a municipal accepted-items benchmark, (2) wishcycles caught per session, and (3) the share of items diverted to non-landfill streams. A campus or building pilot could use QR deep links on bins and bin-contamination audits to test this in the real world.
What it does
Search any household item in a 41-item database with aliases and live suggestions. SortRight returns one of five streams — curbside recycling, compost, landfill, special drop-off, or reuse/donate — and asks a follow-up condition question, such as greasy, broken, still full, food-soiled, or swollen battery, which recomputes the verdict live. Every verdict explains why and names the wishcycle trap. It also includes a resin-code lookup for plastics #1–#7, material fallback heuristics, a session tally of items kept out of landfill and wishcycles caught, and deep links such as /?item=pizza+box for QR codes on bins.
How we built it
SortRight is built around a dependency-free classifier in src/classifier.js (UMD, so the same code runs in the browser and in Node), vanilla HTML, CSS, and JavaScript in public/, and a 30-line zero-dependency Node server in server.js. It uses the Node built-in test runner, with 6 suites passing, and was smoke-tested over live HTTP and in a headless browser.
Challenges we ran into
US curbside rules vary by city, so SortRight encodes mainstream US rules, says so in the UI, and gives a reason for each verdict. A flat item-to-bin table cannot express condition-dependent rules, so conditions are first-class rules in the classifier.
What's next
A city-profile switcher, on-device photo classification feeding the same condition engine, and printable bin-side cheat sheets generated from the database via QR deep links.
Build Disclosure
Build Disclosure: AI-built. Built by AI agents (Tree / Muse fleet) directed by Rene Turcios during the hackathon period; Rene set the direction and reviewed; agents designed, coded, and tested.
Log in or sign up for Devpost to join the conversation.