Custom AI Text Detection: Reduce False Positives by 30%

False positives in academic texts are the main pain point of detection. Student papers with formal style are often mistakenly flagged as AI-generated. In one EdTech case, a client faced 20% false positives on legal documents, leading to mass appeals. We retrained the model on a corpus of legal texts and reduced FP to 4%, applying a combination of statistical methods and a fine-tuned detector. The system processes 50k texts per day with latency <200 ms (p99). Our goal is to adapt the detector to your domain, minimizing false positives while maintaining high accuracy.

How Detectors Differentiate AI Text from Human?

Methods fall into three categories. Statistical methods rely on perplexity and burstiness. Perplexity measures predictability: AI text is smoother. Learn more about perplexity. Burstiness reflects sentence length variation: humans have higher burstiness, AI lower. Fine-tuned detectors, e.g., roberta-base-openai-detector, are trained on GPT samples. The problem: high false positive rates on neutral academic texts. We solve this via augmentation and threshold calibration on your dataset. Watermarking embeds a statistical pattern in tokens during generation—hard to bypass but requires LLM-side support.

How to Improve Detection Accuracy?

Combining methods creates synergy. Fine-tuned detectors yield 85-95% accuracy, 10% higher than statistical on target domains. However, watermarking provides resistance to bypass. We recommend a three-tier approach:

1. Statistical filter (perplexity + burstiness) — quick screening of obvious cases.
2. Fine-tuned model — deep classification with threshold calibration.
3. Watermarking (if supported by generator) — final verification.

On one EdTech project, we reduced false positives from 12% to 3% by adding metadata (generation time, source) and calibrating the perplexity threshold. Result: stable processing of 50k texts per day. Manual content review costs dropped 2.5 times.

Accuracy Comparison by Domain

Fine-tuned models based on RoBERTa consistently boost accuracy by 12-15% over statistical methods on formal-style domains. On medical texts, false positives drop from 20% to 6% after calibration.

What Limitations Should You Consider?

False positive rate of the best detectors is 5–15% on human texts. Academic and legal styles are risk zones. Paraphrasing through another LLM (e.g., LLaMA or Mistral) bypasses most methods. To improve robustness, add generation logs and timestamps — this reduces FP by an additional 2-3%. Detection should be used as one signal, not the sole criterion. Our experience: 5 years in AI solutions, 30+ NLP projects. We guarantee at least a 30% reduction in false positives relative to baseline models.

How We Optimize Performance?

For high-load scenarios, we use quantization (INT8) and ONNX Runtime. vLLM with continuous batching achieves 100+ requests per second on a single GPU (A100). We maintain p99 latency below 200 ms even with batch processing. Model cards document metrics: precision, recall, F1 for each domain. MLOps with MLflow tracks data drift and recalibrates thresholds.

What Our Work Includes

• Analysis of your dataset: collect a representative sample of human and AI texts.
• Model selection and tuning: combination of statistics + fine-tuned model tailored to language and style.
• Deployment: API with latency <200 ms (p99) for streaming processing.
• Documentation and training: we deliver model card and update instructions.
• Support: fix false positives based on your feedback for 3 months.

Order a pilot integration—we’ll evaluate your case in 2 days. Get a consultation on selecting a detection strategy. Contact us—we’ll help find the optimal solution for your budget and tasks.

OpenAI Detector, Kirchenbauer et al. Watermarking LLM

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.