Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

Track: Theme, Rethinking Waste.

What it does

You point your phone at an item, or type its name in Spanish or English. TiraBien tells you which bin it goes in in Malaga, explains the mistake people usually make with that item, and shows the nearest real containers with walking time on a map. It covers the five street bins plus textile, used-oil and battery containers, clean points and pharmacy medicine take-back (73 items, 10 disposal routes).

The waste problem and how success is measured (Theme track paragraph)

Problem targeted: wrong sorting at the household bin. Ecoembes reports that 28.6% of what residents put in the yellow bin in Spain is not packaging (2020), and the same pattern runs through the other streams: drink cartons in the blue bin, receipts and dirty napkins treated as paper, drinking glasses in the glass igloo, plastic non-packaging objects in the yellow bin, batteries and vapes in general waste. Every wrong item contaminates a recyclable batch or starts a fire hazard in collection trucks, and each one is predictable. How we would measure whether it works: (1) the share of yellow-bin content that is non-packaging, from the periodic bin characterisation by Limasam and Ecoembes, comparing neighbourhoods before and after a pilot; (2) the contamination rate of the glass and paper streams at the sorting plants; (3) in the app, how often users override a "likely" answer and which items they search that are missing from the catalog, which shows where residents are confused. Today we have a lab measure only: on 272 real photos of 37 items the bin is correct 96.8% of the time when the app answers confidently (150/155; 79.4% if forced to answer every photo). We have not measured real-world behaviour change.

How we built it

  • On-device AI. SigLIP vision model (ONNX, int8, ~100 MB) in a Web Worker via transformers.js, about 2 s per photo on CPU. Label text embeddings are computed offline, so only the vision tower ships. No server, no account, photos never leave the phone.
  • The AI never decides the bin. It picks which catalog item the photo resembles. A hand-written rule catalog, with sources, decides the bin and explains the mistake. Below a confidence threshold the app asks the user to pick among its top candidates.
  • Real data. 14,400 container locations: Ayuntamiento de Malaga open data (CC BY 4.0) and OpenStreetMap (ODbL), searched with a grid index verified against brute force.
  • Quality. TypeScript, 76 unit tests, a reproducible evaluation script and a public report with its failure cases; typecheck and tests run in CI before the GitHub Pages deploy.

Feasibility (what it takes to deploy for real)

It is a static site: no servers to run, near-zero cost, usable as a web page. Open data and OSM already give the locations. The missing piece is an audit of the rule catalog by the municipal operator (Limasam); the data schema and sources are in the repo so a city could swap in its own rules and containers.

What we would do next

Audit the rules with Limasam; evaluate on phone photos taken in real kitchens and retune the thresholds on a held-out set; add the brown-bin locations the city has not published as open data; add bulky-item pickup booking; extend to other Spanish cities by swapping the rule and container files; collect anonymous, opt-in wrong-answer reports to improve the catalog.

Limitations, disclosed

Malaga only; the rules come from public guidance, not from Limasam's internal data; accuracy was measured on internet photos and the confidence threshold was tuned on the same photos, so 96.8% is optimistic; OpenStreetMap lists only 27 brown bins in the city, so the food-waste map is sparse and the app says so.

Transparency

Designed and written with an AI coding assistant (Claude Code), which also chose the concept on the author's behalf; the author reviewed it and is responsible for it. Work started on 18 Nov 2026, inside the event period. The same project is also entered in SLAY HACKS 2026; Arbiter's rules do not address this.

What we learned

What's next for TiraBien

Built With

Share this project:

Updates

Submission history