Content pipeline for a niche publisher
A specialist publisher wanted a self-hosted content drafting pipeline that respected their editorial voice and their freelance authors.
Their editorial team spent significant effort turning rough freelance drafts into the publisher's voice. They wanted assistance, not replacement, and they would not consider any tool that retained drafts on a third-party server — both for the freelancers' sake and for competitive reasons.
Qwen 2.5 14B on a GPU-equipped VPS plus an n8n workflow with three stages: structural pass, voice pass, and consistency check. Each stage exposes its prompt as a workflow variable so editors can refine the voice over time without touching code. A second workflow ingests the publisher's back catalogue into a Chroma store so the model can ground revisions in past articles.
First-pass editing time on freelance drafts dropped by roughly half. The editor-in-chief described the system as 'a junior editor that has read every issue we have ever published' — a description we have shamelessly borrowed since.
