Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Qwerty AI: Explainable Automated Age Rating and Content Safety Assessment for Russian-Language Screenplays

Nikita Zmanovskii

TL;DR

Qwerty AI introduces a production-ready, end-to-end pipeline for automated age-rating and content safety assessment of Russian-language screenplays under Federal Law 436-FZ. It combines a compact instruction-tuned Phi-3-mini LLM with deterministic rule-based anchors to deliver explainable, legally-grounded ratings at scale, processing 700-page scripts in under 2 minutes on 2×A100 GPUs with 4-bit quantization. The approach leverages LoRA-based fine-tuning, a hybrid loss that includes contrastive terms for document coherence, and a hybrid explainability framework that balances model attributions with explicit legal anchors. Evaluations show 80% exact document-level rating accuracy (MAE 0.27), 80–95% segmentation precision, and strong category-wise detection, while deployment in a hackathon setting demonstrates a practical, self-contained system suitable for production workflows with real editing guidance. The work advances automated content safety in Russian-language NLP by integrating domain-specific legal definitions, long-document processing, and production-ready tooling, and it provides a reproducible methodology and open-source resources to foster further research and deployment.

Abstract

We present Qwerty AI, an end-to-end system for automated age-rating and content-safety assessment of Russian-language screenplays according to Federal Law No. 436-FZ. The system processes full-length scripts (up to 700 pages in under 2 minutes), segments them into narrative units, detects content violations across five categories (violence, sexual content, profanity, substances, frightening elements), and assigns age ratings (0+, 6+, 12+, 16+, 18+) with explainable justifications. Our implementation leverages a fine-tuned Phi-3-mini model with 4-bit quantization, achieving 80% rating accuracy and 80-95% segmentation precision (format-dependent). The system was developed under strict constraints: no external API calls, 80GB VRAM limit, and <5 minute processing time for average scripts. Deployed on Yandex Cloud with CUDA acceleration, Qwerty AI demonstrates practical applicability for production workflows. We achieved these results during the Wink hackathon (November 2025), where our solution addressed real editorial challenges in the Russian media industry.

Qwerty AI: Explainable Automated Age Rating and Content Safety Assessment for Russian-Language Screenplays

TL;DR

Qwerty AI introduces a production-ready, end-to-end pipeline for automated age-rating and content safety assessment of Russian-language screenplays under Federal Law 436-FZ. It combines a compact instruction-tuned Phi-3-mini LLM with deterministic rule-based anchors to deliver explainable, legally-grounded ratings at scale, processing 700-page scripts in under 2 minutes on 2×A100 GPUs with 4-bit quantization. The approach leverages LoRA-based fine-tuning, a hybrid loss that includes contrastive terms for document coherence, and a hybrid explainability framework that balances model attributions with explicit legal anchors. Evaluations show 80% exact document-level rating accuracy (MAE 0.27), 80–95% segmentation precision, and strong category-wise detection, while deployment in a hackathon setting demonstrates a practical, self-contained system suitable for production workflows with real editing guidance. The work advances automated content safety in Russian-language NLP by integrating domain-specific legal definitions, long-document processing, and production-ready tooling, and it provides a reproducible methodology and open-source resources to foster further research and deployment.

Abstract

We present Qwerty AI, an end-to-end system for automated age-rating and content-safety assessment of Russian-language screenplays according to Federal Law No. 436-FZ. The system processes full-length scripts (up to 700 pages in under 2 minutes), segments them into narrative units, detects content violations across five categories (violence, sexual content, profanity, substances, frightening elements), and assigns age ratings (0+, 6+, 12+, 16+, 18+) with explainable justifications. Our implementation leverages a fine-tuned Phi-3-mini model with 4-bit quantization, achieving 80% rating accuracy and 80-95% segmentation precision (format-dependent). The system was developed under strict constraints: no external API calls, 80GB VRAM limit, and <5 minute processing time for average scripts. Deployed on Yandex Cloud with CUDA acceleration, Qwerty AI demonstrates practical applicability for production workflows. We achieved these results during the Wink hackathon (November 2025), where our solution addressed real editorial challenges in the Russian media industry.
Paper Structure (62 sections, 4 equations, 1 figure, 6 tables, 1 algorithm)

This paper contains 62 sections, 4 equations, 1 figure, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Alternative Model Architectures
  4. Content Moderation and Safety
  5. Age Rating and Parental Controls
  6. Explainable NLP
  7. Russian Language NLP
  8. Technical Constraints and Design Decisions
  9. System Architecture
  10. Document Preprocessing
  11. Format Handling
  12. Scene Segmentation
  13. Inference Pipeline
  14. Model Architecture
  15. Prompt Engineering
  16. ...and 47 more sections

Figures (1)

  • Figure 1: Qwerty AI system architecture. The pipeline processes documents through encoding detection, intelligent segmentation, neural analysis, rule-based anchoring, and report generation.