We tried hiring humans to write comments on 20+ Telegram channels. Four teams, $200/week total. It lasted three months before human factor killed it — boredom, inconsistency, no-shows.
So we built an AI system using TDLib (Telegram's official C++ library, not Telethon) that generates contextual comments from 1,000+ unique personas. Each persona has persistent personality traits, opinion drift over time, and natural language quirks including typos and slang. 65-85% of generated comments are threaded replies where personas argue, agree, and reference real-time market data from BingX, CoinGlass, and CoinGecko.
The hardest technical challenge wasn't NLP — it was keeping accounts alive. Telegram's anti-bot systems are aggressive, and we had to build natural typing delays, randomized activity patterns, and session management on TDLib to avoid bans.
We also built ModerAI — a 15-layer anti-spam pipeline with 99.7% accuracy, including voice spam transcription and Vision AI for image spam. Have you tried building anti-spam for Telegram? What approach worked for you?
![Defluffer - reduce token usage 📉 by 45% using this one simple trick! [Earthday challenge]](https://media2.dev.to/dynamic/image/width=1000,height=420,fit=cover,gravity=auto,format=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fiekbgepcutl4jse0sfs0.png)
