Turning an AI writer into a managed service

Cuppa is an AI tool that writes SEO articles. I built its Done-For-You system in about three months: an agency-style workflow where staff produce articles for client teams, with projects, drafts, bulk generation sessions and a separate client view. I also added inline AI suggestions to the editor, article search and tagging, and author profiles.

• Suggestions as accept-or-reject edits

  The model is forced to return pairs of old and new text, and the editor turns each pair into a suggestion the writer accepts or rejects.

  Trade-off Matching paragraphs by text is fragile, but it avoids building a diff engine.

• Prompts in the database

  Prompt templates live in the database with placeholders for tone and language, so admins change them without a deploy.

  Trade-off Prompt changes skip code review, so they need their own care.

• Long work off the request path

  Bulk generation, search processing and publishing run as background functions, not inside page requests.

  Trade-off Users watch progress instead of getting an instant result.