HSAgent

AI platform that automates customs classification with Harmonized System (HS) codes for industrial import-export operations.

Context

AI-driven automated customs classification

HS Agent is an intelligent platform built to automate product classification using Harmonized System (HS) codes. It targets industrial and manufacturing companies that manage complex international import-export workflows. The system analyses technical descriptions, PDF documents, and product data to identify the most accurate customs code. It combines generative AI, semantic search, and rules validation grounded in the HS Explanatory Notes. The reasoning flow mirrors that of an experienced customs specialist—faster, traceable, and consistent. Every decision is explainable: the platform shows which codes were evaluated, ruled out, and confirmed at each stage. The solution runs in a local, containerised environment to protect sensitive business data. HS Agent reduces the risk of misclassification, logistics delays, and compliance issues in a domain that has traditionally relied on scarce manual expertise.

What we did on it...

Vision

Defining the classification problem

The project started from a concrete operational need: customs classification is slow, error-prone, and dependent on rare expertise. The goal was to build an agent that reproduced expert reasoning in a structured, verifiable way.

Structure

Architecture and regulatory logic

The system was designed around a multi-step AI pipeline that combines semantic search, document analysis, and validation against HS Explanatory Notes. Local containerised deployment was chosen to meet enterprise data-security requirements.

Interface

Agent development and user interface

The agent was built with modular components for PDF ingestion, product-description analysis, and code evaluation. The interface guides users step by step through classification, surfacing evaluated and excluded codes to support informed decisions.

Launch

Testing and regulatory validation

The platform was tested on real product datasets, verifying classification accuracy and consistency against official HS Notes. Edge cases—ambiguous items, incomplete descriptions, and multi-chapter products—were handled explicitly before release.