Automating AI Research Delivery — From Papers to Spotify, Substack, Newsletter & YouTube
Tilman Dietrich Every week, hundreds of AI research papers are published. Keeping up is hard. In this talk, I present a fully automated pipeline that scrapes the top 10 papers from Hugging Face's Weekly rankings, generates a ~10-minute two-host podcast episode about the #1 paper, publishes a blog post to Substack, creates YouTube Shorts, and delivers everything as a weekly email newsletter. All of it completely hands-off without any human interference.
The system combines multiple AI APIs in a single pipeline: Claude Opus for generating accessible podcast transcripts and blog posts, Gemini TTS for multi-speaker audio synthesis, and Veo 3.1 for video generation. A key design goal is accessibility — the podcast explains technical jargon, avoids reading math notation aloud, and uses analogies to make cutting-edge research approachable for a broad audience.
I'll walk through the architecture, the Saturday-to-Monday cron-driven workflow, and the practical challenges I solved along the way — from chunking long transcripts for TTS, to automatically publishing a Spotify Podcast and YouTube Videos.