Built a voice-to-blog pipeline for a mate who runs a personal training business. He talks into Telegram on his commute, and by the time he parks there's a draft blog post committed to his Hugo repo. Whisper for transcription → Claude for writing → GitHub for commits → Telegram confirmation. All wired together in n8n, self-hosted on my home server behind Caddy. The only external dependencies are the AI APIs. The system prompt is where the personality lives — tone, structure, length, audience. The infrastructure is generic. Cloned the whole pipeline for a second site in one session. Full writeup: https://blog.dpinkerton.com/posts/voice-to-blog-pipeline/