Content Engine
The publishing pipeline behind this site.
- NLP & information extraction
- Data science & evaluation
- Cloud infrastructure
- Architecture & developer experience
A multi-tenant content service: keyword and SERP research, model-assisted drafting, Postgres storage, and a versioned read API that consuming sites fetch. This site is one tenant of it, and every article on /writing is served by it.
The problem
Publishing that compounds needs research, drafting, editing, images, distribution and cache invalidation working as one pipeline. Assembled by hand from separate tools, it degrades the moment anyone gets busy.
The approach
A content API with tenants, not a CMS. Research and generation sit upstream, a clean read surface sits downstream, and a publish webhook invalidates the consumer’s cache — so the consuming site stays a thin presentation layer. A citation gate holds an article in draft until source coverage clears a threshold, which is an editorial control rather than a disclaimer.
How it works
- A multi-tenant read API, versioned, with per-article citation sources and server-computed reading time
- SERP and keyword research upstream; model-assisted drafting with a citation gate
- Postgres for storage, blob storage for media
- A bearer-authenticated publish webhook driving tag-scoped cache invalidation on the consumer
- Graceful degradation: the consuming site builds green when the API is unreachable, returning an empty list rather than failing
Stack
TypeScript on Node 20, a Hono API on Vercel, Neon Postgres, Vercel Blob, Claude, and scheduled runs via GitHub Actions.
Where it stands
- Twenty-one articles live on /writing, served through the read API
- Replaced an earlier in-repo pipeline, now retired to a redirect manifest so no shared link ever 404s