Skip to content
AI · In production

AI Customs Tariff Classification

An LLM engine that maps products to legally-defensible HSN customs codes — with interactive disambiguation and bulk processing at scale.

All work

The challenge

Customs (HSN) classification is manual, slow, and legally consequential — a wrong code means fines and delays. Teams need answers they can defend, not a black-box guess.

Our approach

  • Hierarchical reasoning that mirrors a customs expert: Section → Chapter → Heading → Subheading, each a distinct LLM step grounded in the General Rules of Interpretation (GRI).
  • Cross-level validation and case-law enrichment so every result is auditable and explainable.
  • Interactive AI-generated questions narrow ambiguous products to the right code, with cached question banks to cut latency and cost.
  • A two-phase, queue-backed bulk pipeline (CSV → questions → classification) on stateless, horizontally-scalable workers.
  • Reliability hardening: per-call timeouts, bounded retries, and env-tunable guards so a slow model can never pin a worker.

The outcome

  • Defensible, GRI-grounded classifications with a full reasoning trail.
  • Bulk classification at scale with versioned history and duplicate detection.
  • Configurable preference modes to steer results toward business rules — only where legally defensible.

Have a problem worth solving with AI?

Tell us what you’re trying to build. We’ll tell you straight whether AI is the right tool — and how we’d ship it.