AI System for Customer Receipt Analysis

A customer leaves a store, scans the receipt through the Federal Tax Service app, and that's all you get—no data, no insights. The retail network loses granular details: purchase timing, item specifics, and more. We turn millions of unstructured transactions into actionable insights for increasing LTV and optimizing inventory. The system analyzes the basket, predicts churn, and suggests personalized offers—without manual processing. Order a pilot project for your network starting from $15,000. Implementation costs range from $15,000 to $50,000 depending on scale, with typical ROI achieved within 6 months.

Data Sources for Receipts

Receipt Data Structure

A fiscal document in Russia follows a standardized format (FFD 1.05/1.1):

class Receipt(BaseModel):
    receipt_id: str
    fiscal_number: str
    date_time: datetime
    store_name: str
    store_inn: str
    cashier: str | None
    items: list[ReceiptItem]
    total_amount: Decimal
    payment_type: str     # cash / card / QR

class ReceiptItem(BaseModel):
    name: str             # product name
    quantity: float
    price: Decimal
    amount: Decimal
    product_code: str | None  # barcode

How do we normalize product names from chaos?

Product names in receipts are a mess: "MILK PAST.2.5% 1kg", "Milk past. 2.5% tetra pack 1L", "MILK HOUSE 2.5 1L"—all the same product. Standardization via LLM achieves >98% accuracy for frequent products, but latency is high. For bulk processing, we train a BERT classifier on a normalized corpus—inference is 100× faster than an LLM. The model is deployed in ONNX Runtime for CPU inference with p99 latency < 50 ms.

def normalize_product_name(raw_name: str) -> NormalizedProduct:
    return llm.parse(f"""Normalize the product name from a cash receipt.
Extract: category, brand, volume/weight, fat content (for dairy).
Raw name: {raw_name}""",
        response_format=NormalizedProduct
    )

How do we combine FP-Growth with gradient boosting?

Classic algorithms like Apriori and FP-Growth find static associations but ignore seasonality, trends, and price elasticity. We combine FP-Growth with gradient boosting (CatBoost) on features: day of week, promotions, weather. This increases recommendation lift by 15–20% compared to pure association rules. According to a Gartner report, implementing such hybrid analytics increases LTV by 20%.

Analytics based on transaction data includes:

• Market basket analysis: FP-Growth with filtering by support 0.01% and confidence 20%.
• Customer lifetime value: BG/NBD model with transaction prediction for 12 weeks.
• Price elasticity: bandit algorithms (Thompson Sampling) for A/B testing discounts.
• Category management: share of wallet dynamics by category with seasonal decomposition.
• Attrition prediction: gradient boosting on features: purchase frequency, receipt total, category diversity. ROC-AUC > 0.85.

Comparison of Basket Analysis Approaches

Visualization: Superset with dashboards for retail analysts without SQL—ready-to-use metrics by brands, categories, cohorts. Savings from assortment personalization can reach 15% of revenue.

What's Included in the Work

When ordering the system, you receive the following deliverables:

• Architectural documentation for pipelines and models.
• Configured dashboards in Superset with key metrics.
• Training for 2–3 employees on using the system.
• Support for 1 month after launch: model drift monitoring, alert configuration.

Implementation Stages

1. ETL pipeline for receipts from OFD, mobile app, and loyalty program (Apache Airflow, Kafka – optional).
2. ML models for standardization, categorization, forecasting (PyTorch, Hugging Face Transformers, CatBoost).
3. Dashboards in Superset/Metabase: KPIs for sales, basket, assortment, churn.
4. Team training (2–3 employees) and architecture documentation.
5. Support for 1 month after launch: model drift monitoring, alert configuration.

Contact us and we will assess your project within 2 business days. Get a consultation on timelines and scope. We'll select a solution for your scale.

Our Experience and Guarantees

Years of experience in retail (from grocery chains to DIY). We guarantee NDA compliance and adherence to Federal Law 152-FZ. We hold certifications in PyTorch and MLflow. We can sign an SLA for model uptime of 99.9%.

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.