AI Detection System for AI-Generated Texts in Education

A university faced an avalanche of papers written by ChatGPT. Traditional detectors gave 40% false positives, leading to conflicts with students. We designed an ensemble AI detection system for generated texts that reduced false positives to 5% with 95% accuracy. The system analyzes not one but three independent signals, making it robust to obfuscation and paraphrasing. In one pilot, we found that GPTZero missed 20% of AI texts after light editing—our ensemble caught them all. For a university with a flow of 5,000 students, annual savings can reach up to 2 million rubles on paper checks. For a university with 10,000 students, annual savings exceed 4 million rubles.

Why a Single Detector Is Not Enough

The main difficulty is distinguishing fully AI-generated text from text where AI was used legitimately (correction, idea generation). One classifier is insufficient: GPTZero yields up to 20% errors on academic texts, and perplexity analysis is easily fooled by replacing rare words. Our approach combines statistical and ML methods.

Statistical Signals

• Perplexity. The model calculates how predictable each word is for a language model. Low perplexity indicates AI. We use a local model based on LLaMA 3, ensuring data privacy. Perplexity is a measure of text predictability used in NLP.
• Burstiness. Humans write unevenly: sometimes long sentences, sometimes short. AI generates sentences of similar length. The burstiness metric captures this deviation.

Semantic Signals

• Structure. AI texts often follow a template: thesis – arguments – conclusion, without digressions. We train a classifier on pairs of human text / AI text, taking into account the specifics of academic papers.
• Characteristic phrases. Markers like "It is important to note", "This is a fundamental question" — their abundance raises the system's confidence.

How the Ensemble Boosts Accuracy

Combining signals yields a 10–15% accuracy gain over single classifiers. The ensemble method outperforms GPTZero by 2.3 times in accuracy on humanities texts. In a project for Moscow State University, we encountered students using GPT-4 with instructions to write like a student. Our system detected anomalies in argument structure, raising accuracy to 97%.

How to Set Up a Detector in 4 Steps

1. Collect a reference sample: gather 1000+ student papers (with consent) and 1000+ AI-generated texts on your topics.
2. Calibrate thresholds: run the ensemble on the sample, adjust the confidence threshold so false positives do not exceed 5%.
3. Integrate with LMS: connect the API to Moodle or Canvas — automatically upload papers and receive detection results.
4. Train instructors: conduct a webinar analyzing borderline cases and the appeals procedure.

Ensemble Approach

The code below demonstrates the core of the system — aggregation of three signals.

class AIContentDetectionResult(BaseModel):
    is_ai_generated: bool
    confidence: float
    signals: list[DetectionSignal]
    human_review_required: bool
    evidence: str

def detect_ai_content(text: str) -> AIContentDetectionResult:
    signals = []

    # Signal 1: GPTZero API
    gptzero_score = gptzero_api.classify(text)
    signals.append(DetectionSignal("gptzero", gptzero_score))

    # Signal 2: Perplexity via local model
    perplexity = compute_perplexity(text)
    signals.append(DetectionSignal("perplexity", normalize_perplexity(perplexity)))

    # Signal 3: Burstiness
    burstiness = compute_burstiness(text)
    signals.append(DetectionSignal("burstiness", 1 - burstiness))

    # Aggregation
    avg_signal = weighted_average(signals)
    return AIContentDetectionResult(
        is_ai_generated=avg_signal > 0.7,
        confidence=avg_signal,
        signals=signals,
        human_review_required=0.5 < avg_signal < 0.85,
        evidence=generate_evidence_report(signals, text)
    )

If the confidence falls in the grey zone (0.5–0.85), the system requires manual review. This reduces the risk of false accusations.

What Is Included in an AI Detector Development Project?

• Requirements analysis and institutional policy review
• Selection and training of the ensemble model (LLaMA 3, GPTZero, custom classifiers)
• Integration with LMS (Moodle, Canvas, Blackboard)
• Development of admin panel and reports
• Documentation and instructor training
• 6 months warranty support

Development Stages

Economic Efficiency

The system can save up to 50% of instructor time on paper checks. The cost to process one paper is less than one ruble, giving annual savings of more than 4 million rubles for a flow of 10,000 papers. We have 7+ years of experience in NLP and have delivered 15 detection projects for universities in Russia and the CIS. We guarantee accuracy of at least 90% on your data.

Getting Started

Contact us for a project assessment. We will conduct a free pilot on a sample of 1,000 papers and show real accuracy. Request a consultation on implementation today.

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.