Cameras that measure what dry-waste sorting centres recover, shift by shift
TL;DR Sortcount puts AI cameras over the belts and reject lines of Indian dry-waste sorting centres. Every shift, it tells the operator what came in, what got recovered and how much value went to the reject heap. Operators pay a monthly fee per line, and recyclers and brands that want traceable material are the second buyer.
India throws out over 1.7 lakh tonnes of waste a day, and an estimated 1.5 million waste pickers do a good chunk of the sorting by hand[1]. Plastic alone is about 34.7 lakh tonnes a year, and roughly 30% of it gets recycled through informal workers[2].
From 1 April 2026, the Solid Waste Management Rules want four-way segregation at source, with reporting on a central CPCB portal and audits[3].
So here's the awkward bit. Most sorting centres can't tell you what share of their dry waste ends up in saleable bales and how much gets dumped as reject. Without that number a city can't judge its contractor and a recycler can't trust its supply.
What it does, in regulatory terms: measurement and reporting only. Sortcount does not issue, buy or sell EPR certificates and is not a registered recycler. Its data supports reports that operators and producers already file[3][4].
A camera and a small edge computer sit over the incoming belt and over the reject stream. They pick out items by material - PET, HDPE, multilayer film, paper, cardboard, metal and glass. The dashboard shows tonnes in, estimated recovery by material and the value lost to reject for every shift. Operators use it to move pickers to where value is slipping through, to grade bales before sale and to show the city real numbers. Brands and producer responsibility organisations can buy a material-flow record for their EPR traceability.
The seed wanted AI robots doing the sorting. I don't buy that for small Indian centres where hand sorting is cheap, and robots would take income away from the workers already there. Indian sorting-robot makers already sell AI and air-jet sorting lines in India anyway[5]. Measuring first is cheaper, keeps the pickers in work and shows you where a machine would actually pay off later.
The first buyer is the centre operator, which could be a municipal contractor, an NGO-run centre or a private aggregator. The test price is a monthly fee per line, and that's my assumption because I found no public prices for sorting analytics in India. The second buyer is a recycler or producer responsibility organisation that wants proof of where material came from. One UK waste-analytics firm sells the same kind of analyser abroad but doesn't publish a price[6].
The 2026 waste rules kicked in on 1 April 2026, with four waste streams, a central portal and mandatory audits[3]. Plastic EPR already runs through CPCB's centralised portal[4]. And one Indian sorting-machine maker reports sorting 1,00,000 tonnes of plastic in FY2024 with its machines[2], so Indian recyclers clearly will pay for machine vision.
Nobody publishes how many material recovery facilities India has or what they spend on sorting analytics. The closest figure is that UK analytics firm's - more than 300 analyser units at over 65 recovery facilities in more than 20 countries[6]. So operators abroad do pay for measurement alone. India's 1.7 lakh tonnes a day[1] is adjacent though. It's total waste, not your market.
An Ahmedabad robotics firm sells a vision system for conveyors and a machine that pairs vision with air jets to sort by polymer and colour[5]. A UK analytics firm sells analyser hardware that sits above belts, with dashboards for operators[6]. A US robotics company builds and runs whole AI sorting plants[7]. Large Indian waste-management platforms say they use AI in their own sorting[1].
What's missing is a cheap, measurement-only product for small Indian centres that sort by hand.
| Difficulty | Moderate (49/100) |
| Starting capital | ₹1–1.8 Cr |
| Time to launch | 9–12 months |
Indian dry waste is wet, dirty and full of multilayer packets, so a model trained on European belts won't transfer and you'll be labelling your own images for months. Dust, bad lighting and power cuts are daily life on a sorting floor. The buyers run on thin margins too, which means the product has to show its value in rupees within weeks.
Sortcount measures and reports, so it needs no waste licence itself. Your customers hold the licences, and you should stay clear of their roles.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
Don't let AI numbers go into EPR filings without spot checks. Soiled, mixed waste can fool a model, and those records carry legal weight.
Want to work out the AI layer for Sortcount? Talk to Yash →
I think this is worth a shot if you have computer vision skills and don't mind spending months on sorting floors. If you want to sell software from an office then this isn't for you. First prove at three centres that the data helps recover at least a few percent more saleable material, enough to cover the monthly fee.
Plan for roughly ₹1–1.8 Cr, much of it on labelling Indian waste images and on rugged camera kits.
No waste licence, since you only measure. You'll need to follow the DPDP Act for workers on camera and keep your records separate from EPR filings.
Yes, and firms abroad already sell analysers to recovery facilities. Check local data and waste reporting rules first.
The free Sortcount validation kit has the line-by-line capital breakdown, the effort scoring, the waste-rule, EPR and worker-privacy checklist, a 14-day validation sprint with pass/fail targets, a 90-day launch plan, 10 interview questions for centre operators and recyclers, and 5 assumptions to test. Download the kit →
Thinking about building Sortcount or something close to it? The make-or-break work is training models on Indian waste and proving rupee gains to operators on thin margins, and that's the kind of thing I help founders work through. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.