VeriKarbon

Track: Carbon Markets & Emissions Transparency. VeriKarbon tackles double countingand slow, opaque verification in CBAM carbon pricing. It also works as a digital productpassport for each shipment.

Live dashboard: https://verikarbon.streamlit.app ·

Code: https://github.com/mafaza1310/verikarbon ·

Video: https://youtu.be/X3k9_xQjQEc

Inspiration Since 1 January 2026, the EU's Carbon Border Adjustment Mechanism (CBAM) chargesimporters for the carbon embedded in steel, cement, aluminium and fertilisers. Exporterswho can't prove their actual emissions pay default values, which are much higher, so acleaner plant gets no reward for being cleaner. Türkiye, host of COP31, is one of the EU'sbiggest suppliers of these goods, and its new national ETS creates a domestic carbonprice that importers may deduct. That raises a new risk: the same Turkish paymentdeducted twice.

What it does: AI integrity layer: checks every hour of a steel plant's monitoring data with EAFphysics rules plus an Isolation Forest. On hours of synthetic data with 158manipulated hours, it caught 97% at 88% precision and reduced the verifier'sworkload to 6% of hours.

CarbonPassport (smart contract): one non-transferable passport per batch.Precursor emissions are inherited from billet to rebar, tonnes can never be used twice,and an operator can never verify its own passport. Only hashes go on-chain, so plantdata stays private.

CarbonPriceLedger (smart contract): a carbon price paid in Türkiye can be deducted only once in total, and never more than the importer actually declared.

Dashboard: CBAM cost with default values vs. verified data, the AI's flagged hours,and passports read directly from the blockchain.

For one 5,000 t rebar shipment, CBAM costs €23,911 with default values vs. €1,548 withverified data in 2026, and €1.13M vs. €62k by 2034.

How we built it: Solidity 0.8.24 with OpenZeppelin AccessControl, Hardhat and 11 tests, deployed onEthereum Sepolia. Python with pandas and scikit-learn for the AI layer, keccak Merkleroots linking the data to the chain, and a Streamlit + Plotly dashboard that reads thecontracts with web3.py.

Reference values are sourced: Türkiye's grid emission factor(0.442 tCO₂/MWh, Ministry of Energy, 2022), IPCC fuel factors, and published EAFconsumption ranges.

Challenges we ran into: Making the AI results honest. Our first test data was too easy and scored near 100%. Wemade the subtler, such as 10–30% electricity under-reporting, and addedlegitimate noise like furnace restarts. Then the ML layer had to earn its place, and it did: itcaught 18 under-reports that the physics rules missed.

Accomplishments that we're proud of: The full story runs on Sepolia: data checked by AI, passport verified, shipment declaredinto the EU, carbon price deducted, and a second deduction rejected on-chain, permanently recorded and checkable by anyone.

What we learned: Blockchain only adds value where several parties distrust each other. CBAM, withexporters, verifiers, EU importers and national registries, is exactly that case.

What's next for VeriKarbon: Pilots with Turkish EAF mills through the chambers of commerce, extension to cementand aluminium, a permissioned consortium chain anchored to a public chain, and hybridpost-quantum (ML-DSA) verifier signatures for records that must be kept for years.

All plant data in this prototype is synthetic and labelled as such.

Built With

• ethereum
• hardhat
• openzeppelin
• pandas
• plotly
• python
• scikit-learn
• solidity
• streamlit
• web3.py