Zero-Shot Profanity Filter: AI moderation for text and images

Overview

Zero-Shot Profanity Filter is an AI moderation tool that detects and filters profane text, checks image safety, and supports Telegram group moderation.

What Users Can Do

• Check Profanity: Check whether input text is profane with confidence scores using a zero-shot classifier.
• Filter Text: Filter and censor profane text using multiple modes (full, word-level/sentence-level, aggressive).
• Upload Images: Upload images to detect NSFW/profane visual content.
• Telegram Bot: Use a Telegram bot that removes profane messages, applies strike rules, and bans repeat offenders.
• Monitor API: Monitor API health and integrate endpoints into moderation workflows.

Why I built this

I built this platform to solve a fundamental issue I faced while studying:

• Unmoderated online spaces often become unsafe due to abusive or profane language.
• Most simple keyword-based filters miss context and multilingual variations.
• Communities also need basic image safety checks, not only text filtering.
• Group moderation should be automated to reduce manual admin effort.

Tech Stack

• Python
• Flask
• Transformers
• PyTorch
• Pillow
• HTML
• CSS
• JavaScript
• python-telegram-bot
• python-dotenv

After launch & Impact

• Built a unified moderation workflow for both text and images in one project.
• Added flexible filtering modes and threshold control to adapt to different safety requirements.
• Enabled practical Telegram moderation with strike-based enforcement and automatic banning.
• Improved understanding of real-time AI-assisted moderation and deployment trade-offs.
• Established a reusable API foundation for integrating moderation into other apps.

Future Plans

• Improve model performance and threshold calibration for better accuracy across different languages.
• Add user-level moderation analytics and moderation history dashboards.
• Add support for custom moderation policies and category-based filtering.
• Optimize inference performance for lower latency and production-scale usage.
• Deploy a hosted version with authentication and usage limits.