My Device Configuration
Twitter/X Content Automation Pipeline
A fully automated, event-driven content pipeline that takes raw news and publishes it to Twitter/X on a schedule, with no manual writing or posting.
Architecture: A Telegram-based scraper ingests news in real time and writes it to Google Sheets, which acts as the pipeline's lightweight data store and trigger source for n8n. From there, an n8n workflow runs the generation stage — producing post ideas and full article drafts — and routes each draft through a one-click Gmail approval gate before it's cleared for publishing.
Publishing itself is decoupled from n8n entirely: a scheduled GitHub Actions workflow (cron-based) handles the actual post to X, using a Codeberg-maintained fork of twikit (chosen after the upstream package threw KEY_BYTE encoding errors in production) and session cookies extracted from an authenticated browser session, stored securely as a GitHub Actions secret rather than relying on X's paid API.
The system includes its own scheduling and rate-limiting logic — a time gate restricting automated posts to 3–4 AM IST and a post-limiter to prevent duplicate/excess posting — both of which can be manually overridden for time-sensitive content. Image generation defaults to Pollinations.ai after reliability issues with HuggingFace's inference endpoints.
This is the second iteration of a Twitter-automation system I'd built previously, which also covered idea generation, DALL·E image generation, and reply-chain posting. This version was rebuilt with a more resilient publishing layer and a human-in-the-loop approval system.
How long did it take you to build?
It took me around 2 weeks to build and get the system working end-to-end.
How much would you charge for creating something like this?
For a system with this level of automation, integrations, custom logic, and reliability handling, I'd charge somewhere in the range of $750–$1,200, depending on the exact requirements and level of customization.
02 Jul 2026
Keywords
AI Usage
AI is used at the content-generation layer: converting scraped news into post ideas, drafting full articles/captions, and generating accompanying images from those drafts. Every architectural and reliability decision — the choice of publishing channel (GitHub Actions over a always-on server), the fork of twikit and the cookie-based auth workaround, the fallback from HuggingFace to Pollinations.ai for image generation, the approval gate before anything is published, and the rate-limiting/scheduling logic — was designed and implemented manually. Image upload to Twitter via this pipeline remains an open engineering problem I'm actively iterating on.
