Sid Mehta Works
Collectio

Collectio

A Panaji waste system that holds uncertain sorting decisions instead of guessing.

  • Role Designer and engineer
  • Client Goa Waste Management
  • Year 2019 to 2026

Project highlights

72.1%

in the lab, on unseen photographs

29%

on a raw crop off a real belt

0

wrong dispatches, 8 of 23 held

Overview

A municipal waste system for Panaji: a driver's route, a sorting camera running a network I trained from scratch, and a ward queue that holds what the model cannot vouch for.

AI · Civic systemsDesigner and engineer · Goa Waste Management

Why I chose waste: India generates 62 million tonnes a year, half of it dumped into landfill sites
For the last five years, no segregation data from the 191 villages was shown to the government

Clinton Vaz runs vRecycle, a waste collector in Goa. We spoke about what collection actually looks like on the ground.

A call with Clinton Vaz, and the podcast and press coverage of his work at vRecycle
Feature, benefit and action mapped for GWMC, collectors and segregators
The information architecture: how GWMC, collectors, segregators and the waste stream connect

The collector's day: confirm a pickup, report an issue, weigh the load, sign off. Nothing typed twice.

A six category classifier, 95,006 parameters, trained from scratch in PyTorch on TrashNet.

Same model, real conveyor footage. Studio photographs are not what a truck sees.

The confusion matrix, the drop from lab to belt, what the model was never taught, and what shipping it costs

Below 55% confidence it stops guessing and asks a person.

The brand, on the street: bins, a kiosk, a truck.

Four Collectio bins, red glass yellow blue for plastic glass cans and paper
A man feeding a bottle into a Collectio recycling kiosk
A Collectio branded collection truck
Collectio feedback: pull quotes from Clinton Vaz of vRecycle and Nicholas of GWMC, and twelve raw notes from both
Reflection

72% is not a good number. That is why the product is built around abstention. A softmax always answers. The product should not. Before anything is sent, the belt asks two more questions, have you seen this before and how sure are you, and fails either one to a person. What is still wrong is named openly. The belt footage is generated reference footage; the tracking and every classification on it are real, but a real camera will bring dirt, glare and overlap the training set never saw. The next version starts with the belt, not the lab.

Back to top
Next project Obin An agent writes the memo. A person checks the numbers. Back to top