Podcast-to-Everything Content Pipeline

What it does

Takes a single podcast episode (latest episode of an RSS feed, or a raw transcript) and runs it through a 7-step pipeline that produces, on average, 15-20 publish-ready content pieces. Flow: (1) Ingest — download audio from RSS and transcribe with timestamped Whisper, or clean a supplied .txt/.srt/.vtt; filler words are stripped for written content while the original timestamps are preserved for clip suggestions. (2) Editorial Brain — segment the transcript into 7 "content atom" types: narrative arcs, quotable moments (screenshot test), controversial takes, data points, stories, frameworks, predictions. (3) Generation — turn each atom into platform-native pieces: 3-5 short-form clips (hook + caption + b-roll), 2-3 Twitter/X threads (280-char rule, each tweet standalone), 1 LinkedIn article, 1 newsletter section, 3-5 quote cards, 1 SEO blog outline, 1 Shorts/TikTok script. (4) Scoring — rate each piece on novelty / controversy / utility. (5) Dedup — cut >70% semantic overlap both within the batch and against the last 30 days, keeping the higher-scored piece. (6) Calendar — slot into a weekly plan using platform peak-hour rules. (7) Output — output.json (agent-readable) + output.html (marketplace visual).

When to use it

For "repurpose this episode", "podcast to social", "content calendar from RSS feed", "extract clips from this episode", or any need to batch-process multiple episodes into a deduplicated editorial calendar. Built for creators and content teams who want a weekly cross-platform distribution plan out of a single audio recording.

Method / frameworks

The core formula is Viral Score = (Novelty × 0.4) + (Controversy × 0.3) + (Utility × 0.3); thresholds are 80+ priority publish, 60-79 calendar fill, <40 cut. Atom extraction and per-platform templates (hook-setup-payoff-CTA short-form structure, thread standalone rule) are preserved from the ericosiu/ai-marketing-skills (MIT) method. Short-form pieces align with industry benchmarks: 70%+ intro retention past the 3-second mark, 30-60s clip length, a 10-20 clips-per-episode ceiling. The dedup engine tracks content hashes in content_history.json to prevent cannibalization.