From January 2026, beverage companies in Korea must use 10% recycled content in their bottles. Yet the only way to prove that percentage is paperwork. We replace it with records measured right on the sorting line.
Feedstock is plentiful. What's missing is a record proving that a given batch is food-grade. Sorting is judged by eye and nobody measures batch composition — so good material cannot be proven, and sells at commodity grade.
We do not replace the sorter. We mount a device above the running conveyor. All hardware is off-the-shelf — no equipment swap, no line stoppage.
Cameras and NIR sensors scan PET flakes on the belt, estimating clear/colored ratio, label residue, and contamination by other plastics (PP·PE·PVC) in real time — turning visual inspection into continuous data.
Input lot → sorting → grinding → washing → shipment is bound to one batch ID, with composition, yield, process conditions and energy logged. Every value is reported with a confidence interval.
Batch records convert automatically into Korea's usage reports, EU data carriers (QR), and California filings. A QR scan traces which batch a bottle's recycled content came from.
AI is used only for detection; regulatory figures are generated by deterministic rules. The device retrofits onto the existing line — the only workflow change is scanning a batch tag.
Not a market trend — statutory deadlines. Records cannot be created retroactively; they must start accumulating now.
Tethered caps mandated — the mixed-material problem remains
10% recycled content mandatory for clear PET bottles
PPWR fully applies
Machine-readable content labels (data carriers) on every package — paperwork will not pass
Korea 30% · EU food-contact PET 30% · California 50%
We start at sorting plants. The fees are small, but the data is created only there — and that data underwrites every brand contract.
Detection unit + batch ledger — monthly subscription per line. Preventing downgrades lifts per-tonne prices; estimated benefit ≈ 3× the fee.
Recycled-content verification reports — annual contract. Missing the mandate means sales restrictions and fines; paperwork alone cannot satisfy the EU.
QR & API integration — annual license. From August 2028 the EU requires machine-readable content data; without it, no market access.
Sorters and analytics AI stop at "measuring"; DPP and mass-balance schemes only "prove" — from paperwork. Mass balance is plant-level arithmetic, and cannot answer the EU's 2028 package-level requirement.
| Capability | Sorters (TOMRA etc.) | Flow-analytics AI (Greyparrot etc.) | DPP · mass balance (Circularise · ISCC) | Flake Ledger |
|---|---|---|---|---|
| Real-time composition measurement on the belt | △ | ○ | × | ○ |
| Batch-level composition & yield records | × | △ | × | ○ |
| Values reported with confidence intervals | × | × | × | ○ |
| Auto-generated KR reports & EU data carriers | × | × | △ | ○ |
| Retrofit onto existing sorting lines | × | ○ | — | ○ |
| Both sorters (upstream) and brands (downstream) as customers | × | × | × | ○ |
The moat is not an algorithm — it is devices installed across sorting plants and the batch records they accumulate. Records cannot be back-dated.
Saved when 1 t of rPET replaces virgin PET (2.15 → 0.45 kgCO₂e/kg, ISO 14044)
Assumes food-grade conversion improving 40%→55% per site · to be replaced by field measurements
B.S. in Computer Science, Georgia Tech.
Ph.D. candidate, Seoul National University AI Graduate School — Computer Vision.
He builds the real-time composition-detection core himself; polymer and recycling-process expertise is covered by advisors (a polymer professor and a reclaimer plant process lead).