Case study
Editorial Pipeline
Automated research → draft → editorial review loop → publish for multiple AI characters with distinct topic interests and writing personas, with citations and posting state tracked end-to-end.
Problem
- Maintaining publishing cadence needs a repeatable workflow: research, outlining, drafting, editing, and posting.
- Quality drops without consistent source collection and citation discipline.
- Multiple character personas with different interests need constraints to prevent tone/topic drift.
- Publishing becomes brittle when posting state and retries aren’t tracked end-to-end.
Solution
- Implemented scheduled + manually triggered runs: research → outline → draft → editor/reviewer loop → final → post.
- Research step performs web research, captures source URLs, and produces structured notes that carry forward.
- Writing step generates an outline first, then expands into a draft that references collected sources.
- Enforced per-character constraints via configuration (topics, posting frequency, custom prompt rules). This is writing persona—not audio/voice synthesis.
- Added a review loop that critiques grounding and citation presence and triggers targeted revisions.
- Persisted drafts, sources, and publish state to enable retries and safe re-runs without duplicate posts.
- Posted to a CMS via API, including scheduled publishing where supported.
Architecture
Results
10+ articles/week
Automated cadence across multiple character personas
Review loop
Editor agent critiques drafts and triggers targeted revisions
Idempotent publishing
Safe re-runs without duplicate posts to CMS
Citation tracking
Source URLs captured during research and carried forward
How this maps to client use cases
- Content ops teams that need an auditable pipeline from research to publish.
- Editorial teams that need multiple voices with a review loop and citation discipline.
- Developer marketing or media sites that need reliable API-based posting with scheduling and retries.