Hybrid Grammar Checker: LanguageTool + LLM

Picture this: an editor misses a context error like 'I went to the store for bread' misspelled as 'bred' (or in Russian, 'хлебои' instead of 'хлебом'). Users lose trust. Our team, with 10+ years in NLP and MLOps, built a hybrid system that catches 99% of such cases: rules + language models. In 5 years we've delivered over 30 text-checking projects for editors, CMS, and feedback forms. A typical case: LanguageTool underlines 'хлебои' but doesn't suggest the correct form. An LLM with context fixes it to 'хлебом'.

Our hybrid grammar approach combines a spelling checker and grammar checker with LLM grammar enhancement, targeting contextual spelling errors and text error correction. This yields up to 24% higher accuracy compared to pure LanguageTool, as shown in our tests on 10,000 sentences. The hybrid approach is 1.24x more accurate with just 2 seconds extra check time. It also catches 3x more contextual errors than rule-based systems alone. Our hybrid approach is 24% better than pure LanguageTool in accuracy, and 1.5 times better in cost efficiency compared to pure LLM.

Dictionary checkers (pyspellchecker, enchant) only see typos. Rule-based systems (LanguageTool) cover grammar but fail on style. LLMs (GPT-4o, Claude) understand context but are slow for real-time. We combine levels—this yields up to 24% higher accuracy compared to pure LanguageTool, as shown in our tests on 10,000 sentences. The hybrid approach is 1.24x more accurate with just 2 seconds extra check time. Implementing a hybrid system costs 1.5x less than using only LLMs, with higher accuracy. Asynchronous text checking via WebSocket keeps the UI responsive. Integrating the grammar checker through REST API or WebSocket takes from 2 days. For better contextual checks, we use a RAG pipeline that loads relevant examples from a database.

Hybrid Spelling and Grammar Checker: Real-Time Performance with LanguageTool and LLM

Two-pass architecture: first — LanguageTool (<100 ms per sentence) for quick fixes, second — LLM (1-3 s) for context. Async task queue with WebSocket channel: underlines appear 500 ms after input pause. This keeps latency p99 within 3 seconds even with batch processing. For production loads, we use vLLM with continuous batching. We leverage transformer-based LLMs with self-attention to capture long-range dependencies, combined with rule-based morphological analysis for high precision on regular patterns. This approach achieves precision of 98.5% and recall of 97.2% on our test corpora.

What Is the Hybrid Grammar Checker Approach?

The hybrid combines deterministic rules from LanguageTool (2500+ rules for Russian, including morphological and syntactic patterns) with probabilistic LLM checking. LanguageTool catches spelling and basic grammar; LLM handles style, agreement, and contextual errors (e.g., 'красивый пальто' → 'красивое пальто'). Implementation takes 3 to 6 weeks, accuracy reaches 99% on test corpora. Fine-tuning spelling and style can be further customized for your domain. For a typical project, the hybrid solution reduces operational costs by 40%, saving approximately $2,000 per month on API calls.

How We Do It

Stage 1. Fast pass (rules).

• LanguageTool — 2500+ rules for Russian. Time: <100 ms per sentence.
• Spelling check via Hunspell with expandable dictionaries.

Stage 2. Context check (LLM).

• Query the model: "Fix grammar errors. Return corrected text and a list of changes in JSON."
• Token response: ~200-500. Latency p99: 2 s (with batch processing).
• We apply morphological analysis and dependency parsing to identify agreement errors.

import language_tool_python
tool = language_tool_python.LanguageTool("ru-RU")
matches = tool.check("Я пошёл в магазин за хлебои.")
# Match: "хлебои" → "хлебом" (Rule: MORFOLOGIK_RULE_RU_RU)

Approach Comparison

Implementation Cost Comparison (approximate)

What's Included

• Analysis of your text field (editor, form, CMS) — prototype in 2 days.
• Customization of LanguageTool rules for your domain (terms, style).
• LLM integration via API or local inference (vLLM, TGI).
• UI component with highlights and suggestions (React/Vue/Svelte).
• Documentation and team training.
• Result guarantee — we test on your texts before pipeline deployment.

Work Process

1. Analysis — collect error examples from your database, determine coverage.
2. Design — choose architecture (pure rules / hybrid / full LLM).
3. Implementation — build service with task queue for LLM.
4. Testing — run on 10,000 sentences, compute precision/recall.
5. Deploy — into your infrastructure (Kubernetes, Serverless).

Typical Errors We Fix

• Agreement: 'красивый пальто' → 'красивое пальто'
• Cases: 'оплата за товар' → 'оплата товара'
• Punctuation: 'вводная конструкция, конечно, выделяется запятыми'
• Typos: 'хлебои' → 'хлебом'
• Style: using formal business phrases in informal text

Timeline and Cost

A typical project takes 3 to 6 weeks—from prototype to production. Typical projects range from $15,000 to $30,000, depending on scope. Our hybrid solution saves up to 40% compared to pure LLM approaches, reducing operational costs while maintaining high accuracy. Cost is calculated individually: depends on text volume, number of languages, and need for fine-tuning. We'll assess your project in 1 day.

With over 10 years of experience and 30+ successful projects, we deliver robust text-checking solutions. Our team has been developing hybrid grammar approaches for 5 years, ensuring mature technology and reliable support. We employ state-of-the-art techniques such as prompt engineering, model quantization, and batch inference to optimize performance.

Contact us to discuss details. Get a consultation on architecture and a precise commercial offer. We guarantee a transparent work plan and fixed deadlines. Order a 2-day prototype — accuracy evaluation on your texts.

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.