ClassifyAI - Tariff classification
Automated HS code classification in 60 seconds. 1M+ BTI rulings, GRI justification.
ClassifyAI is The Trade Hub's automated tariff classification tool. From a text description or images, it identifies the applicable HS (Harmonized System) code by drawing on over one million EU BTI/BOI rulings and the complete TARIC nomenclature.
What ClassifyAI does
ClassifyAI takes a product description as input - free text, technical specifications, images - and returns a ranked list of HS codes ordered by confidence, each accompanied by regulatory justification based on the 6 General Interpretive Rules (GIRs).
Average response time is 60 seconds for a full classification.
Who ClassifyAI is for
| Profile | Use case |
|---|---|
| Customs declarants | Quick classification check before filing a declaration |
| Freight forwarders | High-volume processing with batch mode |
| Importers / Exporters | HS code identification to estimate customs duties |
| Compliance officers | Audit of existing classifications, compliance documentation |
| Specialized lawyers | BTI case law research for litigation |
Classification pipeline
ClassifyAI relies on a RAG (Retrieval-Augmented Generation) pipeline in 4 stages:
1. Embedding and hybrid search
The product description is transformed into a semantic vector, then searched against our knowledge base containing:
- 1,000,000+ BTI/BOI rulings from the European Union
- The complete TARIC nomenclature (25,000+ 10-digit codes)
- Section, Chapter and Subheading Notes
The search combines vector similarity and exact text matching to maximize relevance.
2. Reranking
Candidate results are re-ranked by a reranking model to isolate the BTI rulings and codes most relevant to the provided description.
3. Consolidation and reasoning
A reasoning model analyzes the results, the nomenclature hierarchy and applicable legal Notes to produce:
- A ranking of HS codes ordered by confidence score
- The GIR justification for each proposed code (which rule applies and why)
- The complete hierarchy (section, chapter, heading, subheading)
4. Interactive clarification
In interactive mode, if the information provided is insufficient for a reliable classification, ClassifyAI asks targeted clarification questions (constituent material, primary function, intended use) before finalizing the classification.
Clarification questions do not consume classification quota. Only the final classification is counted.
Results
Each classification returns:
| Element | Description |
|---|---|
| Rankings | List of HS codes ordered by confidence (score 0-100) |
| HS code | 6, 8 or 10-digit code depending on achievable precision |
| GIR justification | General Interpretive Rule applied (GIR 1 through 6) |
| Hierarchy | Section, chapter, heading, subheading with descriptions |
| Consolidated description | Normalized reformulation of the product |
| Explanatory text | Detailed reasoning, BTI rulings consulted, TARIC measures |
Usage modes
Interactive mode (chat)
The ClassifyAI web interface operates as a real-time conversation (SSE). The user describes their product, ClassifyAI asks clarification questions if needed, then returns the classification.
This mode is accessible directly from The Trade Hub application, under the ClassifyAI section.
Async mode (jobs)
For programmatic integrations, ClassifyAI exposes an asynchronous pattern based on jobs: job creation, status polling, results retrieval.
Batch mode
Batch mode allows classifying up to 100 products in a single request. Each product is processed independently and results are aggregated in a single response.
Supported languages
ClassifyAI operates in French, English and Spanish. The response language is configurable independently of the input description language.
Knowledge base
| Source | Volume | Update frequency |
|---|---|---|
| BTI/BOI rulings (EBTI-3) | 1,000,000+ rulings | Weekly |
| TARIC nomenclature | 25,000+ 10-digit codes | Daily |
| Section and Chapter Notes | 21 sections, 99 chapters | At each HS revision |
| General Interpretive Rules | 6 GIRs + Explanatory Notes | Stable (HS 2022 revision) |
Last updated