KUL03 - BunkerGuard AI: Verify Every Drop Before You Sign

The $132B marine fuel industry still runs on a piece of paper nobody verifies. Every day, ships burn fuel they didn't measure. Barges pump air into volume readings. Chief Engineers sign Bunker Delivery Notes under pressure — and fraud only surfaces weeks later, when the barge is long gone. We built the fix. BunkerGuard is a multi-agent AI system that transforms a multi-day post-hoc dispute into a 90-second pre-sign-off verdict. Five specialized agents — Surveyor, Investigator, Compliance, Decision, and Chief Engineer — work in concert to ingest live MFM telemetry, run 26 anomaly detection rules, cross-check MARPOL and ISO 8217 compliance, and produce a tamper-evident, auditable evidence package before anyone puts pen to paper. The stack is live. Bedrock Claude Sonnet 4.6 · AWS us-west-2 · Exa live intelligence · Vercel AI Gateway · OpenRouter fallback. This isn't a prototype. It's deployed, it's running, and it's already catching discrepancies that cost operators millions. We're looking for teammates who want to build at the intersection of:  ???? Maritime operations & real-world physical infrastructure ???? Agentic AI with real stakes and real consequences ⚖️ Compliance, fraud detection, and financial integrity

To develop a live AI verification platform — a web application where Chief Engineers upload or photograph a Bunker Delivery Note, and five specialized AI agents instantly analyze it for fraud, anomalies, and compliance violations, returning a signed evidence package and go/no-go verdict before sign-off.

Full-stack developer (Next.js / React / APIs) · AI/LLM engineer (agent orchestration, prompt engineering, Bedrock/Claude) · Maritime domain expert (bunkering operations, MARPOL, BDN workflows) · UX designer (dashboard, evidence UI, human-in-the-loop flows) · Data analyst (anomaly detection, risk scoring, telemetry interpretation)