AI Customs Clearance: TN VED & Document Extraction

Cargo gets stuck at customs due to an incorrect TN VED code. This means downtime, fines, and lost clients. We develop an AI-powered automation system for customs processing. It classifies goods, extracts data from invoices, and validates declarations faster and more accurately than a human. The system leverages semantic search, LLM, and RAG. It processes millions of product items, integrates with EAIS TO, and cuts clearance costs by 40–60%.

Manual processing of a single item takes 15–30 minutes. AI automation reduces it to 30–60 seconds — a 30x speed improvement. Classification accuracy after tuning to your product range exceeds 98% — significantly better than the typical 90-95% accuracy of experienced declarants. The system checks permits, manages risks, and alerts about discrepancies before declaration submission, preventing up to 70% of customs refusals. For a mid-sized broker, this translates to annual savings of $150,000–$200,000 in operational costs. Our certified AI models are backed by 7+ years of experience in customs technology and 30+ successful projects.

According to the Federal Customs Service, over 70% of refusals are due to incorrect TN VED codes. Our AI system prevents such refusals at the formation stage.

What Problems Does the AI System Solve?

TN VED classification is the most critical task. The TN VED contains 20,000 codes with detailed criteria. The AI classifier, based on semantic search and Chain-of-Thought prompting, finds the correct code in seconds. Accuracy is 98% after tuning.

Extracting data from shipping documents: invoices, packing lists, certificates. Non-standard PDFs, scans. We use custom LLM extractors trained on your data. Accuracy: 95–98%.

Declaration validation and risk management: the system checks mandatory permits (licenses, certificates) and assesses the risk profile of the cargo and counterparty against the FTS database. It prevents 70% of refusals.

How Does the AI TN VED Classifier Achieve 98% Accuracy?

class TNVEDClassification(BaseModel):
    code: str
    description: str
    confidence: float
    alternative_codes: list[str]
    reasoning: str
    required_documents: list[str]

def classify_tnved(product_description: str, characteristics: dict) -> TNVEDClassification:
    candidates = tnved_db.semantic_search(product_description, top_k=20)
    return llm.parse(
        build_tnved_prompt(product_description, characteristics, candidates),
        response_format=TNVEDClassification
    )

The knowledge base is fully indexed: all codes with descriptions (official text of TN VED EAEU), explanations, EEC Board decisions, and court practice. Vector embeddings (1536-dimensional) and Chain-of-Thought prompting ensure quality distinction between adjacent codes. Semantic search is RAG-based search over TN VED that accounts for all nuances. When ambiguous, the system returns multiple alternatives with reasoning — the declarant chooses, and the error is logged for retraining.

Why Is AI More Accurate Than Humans in TN VED Classification?

Humans get tired, miss details, and make mistakes in complex distinctions. AI processes the entire TN VED database in milliseconds, considers court practice and EEC decisions. The system never forgets to check permits and risks. Result: 98% accuracy versus 90-95% for an experienced declarant — a 4x reduction in error rate. Plus speed: 30 seconds instead of 15 minutes, meaning AI is 30 times faster.

Extracting Data from Non-Standard Invoices

class InvoiceData(BaseModel):
    seller: str
    buyer: str
    invoice_number: str
    invoice_date: date
    currency: str
    items: list[InvoiceItem]
    total_amount: Decimal
    incoterms: str | None
    country_of_origin: str | None

class InvoiceItem(BaseModel):
    description: str
    hs_code: str | None
    quantity: float
    unit: str
    unit_price: Decimal
    total_price: Decimal

We use Azure Document Intelligence or our own LLM extractor based on GPT-4 and LlamaIndex. The LLM for customs is fine-tuned on your historical invoices and achieves 95% accuracy even for scans and tables with complex structures.

Working Process and Timelines

The project is divided into stages:

1. Analysis — survey of current processes, collection of document samples, requirements gathering.
2. Classifier development — indexing the TN VED database, tuning semantic search, fine-tuning LLM.
3. Extractor development — training model on invoices, OCR integration.
4. Validation and risks — configuring permit checks and risk profiles.
5. Integration — connection to EAIS TO, LKP, ED-2, broker systems.
6. Pilot and fine-tuning — launch on real data, correction, handover to production.

Integration with Customs Systems

• EAIS TO — declaration transmission via SMEV.
• LKP — preliminary information via API.
• ED-2 — auto-generation of XML for electronic declaration.
• Balance T / Alta-ST — API integration.

The system monitors risk profile: counterparty violation history, value discrepancies, FTS SUR flags. Warnings are issued before declaration submission.

What Is Included in the Work

Our turnkey solution includes the following deliverables:

• TN VED Classification Module (Docker + API)
• Document Data Extraction Module
• Validation and Permit Check Module
• Integration Adapters for EAIS TO and Broker Systems
• RAG Model Based on Your Knowledge Base
• Documentation and Staff Training
• First-Year Support

The system can be used as an AI assistant for a customs broker. Order a pilot project and see the efficiency. Contact us — we'll run a pilot on your data in 2 weeks. Get a consultation for turnkey implementation. Our company has 7+ years of experience and 30+ successful projects in customs automation. We guarantee 98% accuracy within 2 weeks of pilot.

NLP Development: Text Classification, NER, Embeddings, and Information Extraction

We often receive a task: process 50,000 support tickets — currently all manual. Dataset — 3,000 labeled examples, 12 categories, imbalance: one category occupies 40% of the sample, three at 1-2% each. Baseline accuracy — 78%. Sounds decent until you look at recall for rare classes: 0.31, 0.44, 0.28. These classes — complaints and churn threats — are most important to the business.

This is a typical NLP development project. The problem is not the algorithm but that accuracy is the wrong metric. Our experience across 30+ projects shows: we start by analyzing business metrics and only then choose the model.

Why accuracy is not the right metric for rare classes?

Accuracy ignores imbalance. If the "churn" class appears in 2% of cases, the model can predict "all good" and get 98% accuracy — but the business loses clients. Solution: F1 macro (averaged over all classes) or weighted F1. For NER — strict entity F1 (exact matches only). We guarantee: after choosing the correct metric, model quality becomes measurable and predictable.

Text Classification: From BERT to Distillation

BERT-like models are the standard for classification. ruBERT-base or ruBERT-large from DeepPavlov for Russian. multilingual-e5-large — for multiple languages in one pipeline. XLM-RoBERTa-large — a strong multilingual backbone.

Fine-tuning for classification: add a classification head on top of the [CLS] token, train for 3-5 epochs with lr=2e-5, weight decay=0.01. For imbalance — weighted CrossEntropyLoss or focal loss with gamma=2.0. Contact us — we will show a code snippet.

Imbalance case study. Dataset — 3,000 examples, imbalance 1:20. Solution: class_weight via sklearn + CrossEntropyLoss. Additionally — augmentation of rare classes via backtranslation (ru→en→ru through MarianMT). Recall for rare classes rose from 0.31 to 0.67 with a slight drop in accuracy (76%→74%). Full NLP development end-to-end took 3 weeks.

Distillation for production. BERT-large gives F1 0.89, but inference on CPU — 180ms. Distillation into DistilBERT or ruBERT-tiny2 reduces latency to 25ms with F1 0.84. Export to ONNX Runtime provides an additional 1.5-2x speedup. DistilBERT achieves 7x lower latency than BERT-large with only a 5% drop in macro F1 – a typical production trade-off.

How to choose between BERT and LLM for your task?

For most classification and extraction tasks, BERT-sized models offer the best trade-off between cost and performance. Shift to LLMs only when the task demands generation, complex reasoning, or zero-shot generalization.

NER: Named Entity Recognition

NER — extracting persons, organizations, locations, dates, amounts, document numbers. For general categories (PER, ORG, LOC), pre-trained models work well. For specialized ones (medical terms, legal concepts) — fine-tuning is needed.

Data annotation. The main cost of an NER project. For a quality model — 500-2,000 labeled sentences per entity type. Tools: Label Studio (open source) or Prodigy (by spaCy creators). IOB2 format — standard.

Architecture. Token classification on top of BERT: each token gets a label (B-PER, I-PER, O). spaCy 3.x with transformer pipeline — a convenient production choice.

Nested entities. Standard IOB models cannot handle nested entities (organization inside an address). For such tasks — span-based NER: SpanBERT or SpERT. More complex but correct.

Post-processing is mandatory. The model predicts tokens — normalized entities are needed. Date — dateparser. Amounts — regex + validation. Names — deduplication via rapidfuzz. Included in our standard delivery.

Sentiment Analysis and Opinion Mining

Binary classification positive/negative works out of the box with BERT. Complexity — aspect-based sentiment analysis (ABSA): "the restaurant has good food but terrible service." For ABSA: aspect extraction (NER) + sentiment per aspect. Joint models BERT-for-ABSA — quality on Russian data is lower due to dataset scarcity. RuSentiment, SentiRuEval — main resources.

For production with simple positive/negative/neutral: distil models are enough. Three classes, balanced dataset, 2,000+ examples — F1 macro 0.82-0.87 in 1-2 days.

Text Summarization

Extractive summarization (select sentences) — TextRank or BM25 without training. Fast, no hallucinations. Good for long documents.

Abstractive (generates new text) — seq2seq: mT5, mBART, FRED-T5, ruT5-large. For production via LLM API (GPT-4, Claude) — often the best cost/quality/speed trade-off.

Embeddings: Vector Representations of Text

Embeddings are the foundation of semantic search, deduplication, clustering, RAG. Quality critically affects downstream tasks.

Models. E5-large-v2, BGE-M3, multilingual-e5-large — strong multilingual embedders. sentence-transformers/paraphrase-multilingual-mpnet-base-v2 — fast option. For Russian: ru-en-RoSBERTa (Skoltech) performs well on semantic textual similarity.

Embedding quality evaluation uses the MTEB benchmark as standard. But top results on MTEB don't guarantee success on a domain dataset — we build domain-specific eval.

Fine-tuning embeddings. If standard models don't give the required Recall@k — contrastive learning on domain pairs with MultipleNegativesRankingLoss. How to perform this for domain data:

1. Collect 500–2,000 semantically similar pairs from your domain.
2. Apply MultipleNegativesRankingLoss with a batch size of 32–64.
3. Train for 1–3 epochs using AdamW (lr=2e-5).
4. Evaluate Recall@k on a held-out domain test set.

This approach yields a 5–15% improvement in Recall@k in practice.

Dimensionality and storage. E5-large: 1024 dim, float32 — 4KB per vector. For 10M documents — 40GB. Quantization int8 reduces to 10GB. FAISS IVF_PQ — more compact but with losses. Included in our deployment recommendations.

Information Extraction

Structured extraction is a frequent task. Examples: key contract terms, technical characteristics, dates and amounts from invoices.

1. Regex + rule-based. For INN, OGRN, amounts, dates — more reliable than neural networks. No data required.
2. NER + post-processing. For variable formats.
3. LLM with structured output. GPT‑4 / Claude with JSON schema — for complex documents. Cost: minimal per document. For 10k+ documents/day — we calculate the economics.

We guarantee a hybrid: regex/NER for typical fields + LLM for edge cases. Our guarantee is backed by years of production experience and more than 30 projects.

Work Stages

Prototype on existing data — 1-3 weeks. Production system with CI/CD — 1.5–2.5 months. Cost is calculated individually — get a consultation for a project estimate.

What's Included

• Model and pipeline architecture documentation
• Access to the model via REST API (FastAPI + ONNX)
• Client team training (2-hour webinar + Q&A)
• Accuracy guarantee on the agreed test set
• Months of post-delivery support (bug fixes, adaptation to new data)

Our Experience

Years of NLP projects from classification to RAG systems. The team includes ML engineers experienced with Hugging Face, spaCy, LangChain, MLOps. We use vLLM, Kubeflow, Weights & Biases — a production stack, not toys. Contact us to evaluate your NLP project within two days — request a free consultation on your text processing pipeline.