AI-Driven Carbon Arbitrage & Edge Verification Engine

Real proof from factory floors via WhatsApp. AI + human verification stops carbon tax penalties, optimizes margins, unlocks green loans.

Comment

Inspiration

With Malaysia's carbon tax launching in 2026 (starting with iron, steel, and energy sectors) and the EU's CBAM imposing punitive tariffs on high-emission exports like steel, aluminum, and palm oil, Malaysian exporters and heavy industries face massive profit erosion—potentially millions in extra costs at borders or through domestic taxes. Traditional carbon tools rely on guesses from spreadsheets, expensive hardware, or slow consulting reports, leaving Scope 3 supply chains (thousands of SMEs) opaque and vulnerable.

I was thinking to build something that could address the initially 'inaddressable', which is the Scope 3 supply chains emmisions. It is meant to be a zero-friction, real-world-proof system that captures physical truth from factory floors and logistics via everyday tools like WhatsApp, verifies it rigorously (AI + human experts), and turns it into immediate financial defense—protecting margins, dodging penalties, optimizing operations, and unlocking green financing.

What it does

CarbonAI is a real-time financial shield and arbitrage engine for carbon-regulated industries. It:

  • Lets factory workers, drivers, and SME suppliers submit raw evidence (photos of receipts, short videos of machinery, voice notes) via WhatsApp—no new apps needed.

  • Uses AI (multimodal analysis) to extract exact emissions data (fuel type/volume, energy source, timestamps) while ignoring noise.

  • Applies ironclad verification with always-on human intervention to check every critical item (receipts, machine conditions, audios, videos) for 100% accuracy and anti-fraud protection (GPS checks, liveness prompts, audio-visual sync).

  • Stores verified data immutably in a secure vault for audit-proof compliance (ISO 14064, Malaysian customs, EU regulators).

  • Executes arbitrage: auto-attaches proof to shipments to avoid CBAM "dirty" tariffs, optimizes factory schedules via MES integration for low-tax energy windows, and feeds banks verified SME data for cheaper green loans (with us earning a spread).

How I built it (Technology involved)

I built CarbonAI as a lean, hackathon-prototype stack focused on edge capture, AI triage, human verification, and execution:

  • Frontend/Ingestion: WhatsApp Business API for zero-friction input (photos, videos, voice notes tied to GPS/phone metadata).
  • AI Backend: Gemini (multimodal model) for extraction and initial confidence scoring of variables from unstructured data.
  • Verification Layer: Custom human-in-the-loop workflow (flags <98.5% confidence + spot-checks) to manually verify uploads, enforce liveness (random prompts), and detect forgery via metadata/audio-visual continuity.
  • Storage & Security: Cryptographic hashing into a private vault for immutable, trade-secret-safe records.
  • Arbitrage Endpoints: Mock APIs/simulations for MES integration, customs document generation, and bank data sharing.

I prioritized ruthless simplicity: no hardware, no complex dashboards for SMEs—just WhatsApp + AI + human rigor.

Challenges I ran into and will possible face

  • Ensuring 100% deterministic accuracy for regulatory use—AI hallucinations could create massive liability, so we had to enforce strict human verification on every low-confidence or critical case without killing speed.

  • Anti-fraud in chaotic supply chains—suppliers might spoof data, so we layered in liveness detection, metadata forcing, and economic deterrence (blacklisting).

  • Scope 3 adoption friction—convincing 5,000+ SMEs to participate without incentives; solved by making input effortless (WhatsApp) and tying value to big-company pull (cheaper loans, contract security).

  • Simulating real integrations (MES, customs APIs, banks) in a hackathon timeframe—we used mocks and focused on core flow proof-of-concept.

  • Balancing AI speed with human oversight—tuning confidence thresholds to minimize human load while guaranteeing trust.

Accomplishments that I'm proud of

  • Built a illustrative prototype that ingests real WhatsApp-style data, analyzes it multimodally, routes for human verification, and outputs verifiable compliance packets + arbitrage triggers.

What I learned

  • Doing hackathon solo is quite mentally-exhausting :')

What's next for AI-Driven Carbon Arbitrage & Edge Verification Engine

  • Pilot with 1–2 apex Malaysian exporters (e.g., steel/aluminum firms on Bursa Malaysia) and a Tier-1 bank (Maybank/CIMB) to validate real tax savings and green loan unlocks.
  • Expand human verification team and automate more triage while keeping 100% oversight for critical paths.
  • Full integrations: Real MES APIs (for production scheduling), Malaysian customs endpoints, bank green-finance gateways.

Built With

Share this project:

Updates