The content factory operates as a Python-based system using Telegram as its primary interface. It monitors 13 news channels, processing 4,299 news items to date. The system uses a multi-step filtering process, including MD5 hashing and fuzzy matching, to eliminate duplicates. An AI model evaluates each news item on a scale of 0 to 100, automatically filtering out more than half of the incoming stream.
The workflow involves an AI agent that drafts posts based on the user's style, performs fact-checking, and generates content versions for Instagram, Facebook, Threads, YouTube, and a website. The system supports voice input for comments via OpenAI Whisper. Each post undergoes a verification process where the model identifies potential discrepancies, though the user retains final approval authority.
The architecture relies on a single process using asyncio to manage tasks, including news scouting and content generation. It integrates various APIs such as Meta Graph, YouTube Data, and Supabase. The system maintains state in JSON and JSONL files rather than a traditional database. Between April 18 and October 1, 2026, the factory produced 83 posts across three channels.
The developer identified several limitations, including the sequential processing of AI requests, which can lead to queues lasting tens of minutes. Other challenges include the lack of an idempotency key for publications and the potential for similar posts when different sources cover the same event. Future improvements aim to address fact-checking robustness and concurrency.
