89lab content pipelines

The problem

Teaching AI in public means posting every day: finding the topic, checking the facts, writing, voiceover, captions, rendering and publishing on several platforms. Done by hand, a day gets missed, a topic repeats or a weak post goes out.

What I built

I built and run each channel as its own pipeline with its own schedule. AI news shorts and AI concept explainers go to the 89lab Facebook page. A trending GitHub repo becomes a YouTube Short and a Threads post, and a build with AI series shows how to make things step by step. Practical AI lessons go to Instagram. A daily learning pipeline researches one AI concept and adds it to a growing library that the explainers draw from. Together they teach what is new, which tools are worth using and how AI agents are designed and built.

How it works

1. Pick a fresh topic: a news story, a trending repo or an AI concept, never one already posted
2. Research it and write a short script that passes length and source checks
3. Voice it, caption it and render the video
4. Publish on schedule, record the post and send me a report on Telegram

Under the hood

Each pipeline is a folder of numbered stages run by a shell script on a schedule. The script, not a model, decides the order, retries and reporting, so model calls are few and bounded. A ledger per pipeline prevents repeats, and checks stop a script that is too long, overhyped or not supported by its source. Local Kokoro speech, timed captions, Remotion and FFmpeg make the videos. When YouTube’s daily upload limit is reached, the finished video goes to Telegram for me to post by hand. A nightly report tracks what the models cost.

What this page does not claim

This page explains the pipelines without running them. No posts are published from the portfolio, no account credentials are exposed and no follower, view or growth numbers are claimed. Some series wait for my approval before they publish.