CN/TARIC code suggestions
The system indicates possible tariff classifications based on product description, technical features, TARIC structure and similar decisions.
AI-assisted customs classification
TaricAI helps find possible CN/TARIC codes, similar BTI/WIT decisions and source documents faster.
The assistant analyzes product descriptions, compares them with BTI/WIT decisions issued by EU customs administrations, uses the TARIC tree and presents sources that help the user independently review classification suggestions.
Important: TaricAI does not issue binding tariff information and does not replace a customs expert. The system suggests possible classifications, source documents and document checklists for independent review before filing a declaration. Similar public BTI/WIT decisions are comparative material and do not constitute a binding decision for the user.
What is the project?
An assistant for customs professionals: classification, sources and declaration documents.
Similar BTI/WIT decisions
TaricAI searches real classification decisions issued by customs administrations of EU Member States.
Declaration document checklist
After a code is selected, the system can use ISZTAR information to indicate documents worth preparing for the customs declaration: certificates, permits, invoices, specifications or other required attachments.
Knowledge sources
The project is based on EU customs data, not model guessing.
Integrated Tariff of the European Union
TARIC is the integrated tariff of the European Union. TaricAI uses the code structure, nomenclature descriptions and relationships between groups of goods. ISZTAR data may also help determine which documents should be attached after the product is classified.
European Binding Tariff Information
EBTI contains BTI/WIT decisions issued by EU customs administrations. Decisions include goods description, code, reasoning and validity period. In TaricAI they are treated as comparative material, not as binding decisions for another case.
Classification decisions
Validation identified an EBTI export covering more than 120,000 decisions. These real cases can power semantic search.
Sources and limitations
TaricAI should show the basis of each suggestion, not replace a customs authority decision.
Official sources
The system is intended to use TARIC, ISZTAR and public EBTI/BTI/WIT decisions. These sources support finding similar cases, codes and document requirements.
Comparative material
Similar BTI/WIT decisions help understand how comparable goods were classified, but they are not binding decisions for the user or the user's specific case.
Documents after classification
Document requirements depend on the code, date, trade direction and applicable measures. The document checklist therefore always requires final review.
How it works
From product description to code, sources and documents.
TaricAI compares goods descriptions with BTI/WIT decisions and TARIC structure. The result is not a certain code, but a list of likely classifications, sources and documents to review before preparing the declaration.
1
The user describes product, material, use and technical features.
2
The system searches similar BTI/WIT decisions and relevant TARIC entries.
3
AI organizes suggestions and identifies missing information.
4
The user reviews sources, selects the appropriate code and receives a document checklist for the declaration.
Security and independence
The system can work with public AI models or with models deployed inside the company.
TaricAI is designed not to depend on a single AI provider. It can use external AI services such as OpenAI, Anthropic or Google, but it can also work with models deployed on customer infrastructure.
Data protection
Some organizations cannot send data to external AI services due to business secrecy, client requirements or security policies.
Provider independence
Model availability can depend on licensing changes, business decisions, government regulations or export restrictions.
Same knowledge base
Regardless of the selected model, the system uses the same sources: TARIC, EBTI/WIT and documents used during analysis.
Why we discuss technology
Architecture is part of the product promise: sources should be verifiable and answers repeatable.
In a finished product, the user mainly cares about the outcome: finding similar decisions, possible codes and source documents faster. At this stage, however, TaricAI is still in the concept, analysis and validation phase. That is why technical choices are described here — they will determine whether the system is trustworthy, repeatable and maintainable.
This is also relevant for people who may want to join the project. We want to show clearly that TaricAI is not intended to be a simple chatbot, but a system based on controlled sources: TARIC, EBTI/WIT, a local database, semantic index and documents that users can independently review.
Agentic flows are possible: the model can plan steps, select tools and decide what to search next. That approach can be flexible, but in formal domains it is harder to control. RAG fits customs classification better because it gives the model room for analysis, while keeping it inside selected sources, data collections, retrieval traces and a clearly defined process.
PostgreSQLsource of truth for codes, decisions, import metadata and synchronization history
Weaviatesemantic index for finding similar goods descriptions and BTI/WIT decisions
EBTI CSVbulk import of BTI/WIT decisions using Issued Since and Download Search Result
TARIC / ISZTARtariff code tree, nomenclature descriptions and document requirements after classification
LLM APIanalysis layer: organizing suggestions, identifying missing information and preparing responses
Source documentslinks to decisions, PDFs and materials that the user can independently review
Cooperation
TaricAI is being developed as an independent early-stage product project.
We are looking for conversations with people and organizations working daily with customs classification, import, export or tariff data. Practical feedback from people who can judge whether the system output would be useful in real work is especially valuable.
People supporting domain validation may receive early access to the test version and have real influence on the product direction.
Contact: info@taricai.com
Roland Rusiecki
Originator of the TaricAI concept. Senior Software Developer experienced in .NET systems, databases and AI/RAG solutions.
Customs experts
For people who want to help evaluate classification suggestions and the usefulness of the system in day-to-day work.
Pilot partners
For companies interested in early access, testing on real cases and shaping product features.
Project blog
Notes from building TaricAI
Short posts about what we discover while building the project: data sources, importers, RAG, AI limitations and product decisions.
Why TaricAI?
Tariff classification requires source work, similar cases and product details. AI can reduce search time, but it cannot pretend to be an official decision.
Read moreEBTI as a knowledge base
BTI/WIT decisions contain goods descriptions, codes and classification reasoning. This is real source material for semantic search.
Read moreAI does not replace the expert
TaricAI should show suggestions, similar decisions and sources. The final decision remains with the user or customs specialist.
Read moreContact
The project is at the concept and validation stage.
We are looking for conversations with customs agencies, importers, exporters and tariff classification specialists.
info@taricai.com